ISSN 1650-674X

TRITA-ETS-2003-3

Induction Bearings
A Homopolar Concept
for High Speed Machines

7RUEM|UQ$/HPENH

Royal Institute of Technology

Department of Electrical Engineering
Electrical Machines and Power Electronics

Stockholm 2003


ISSN 1650-674X
TRITA-ETS-2003-3
 


Abstract
A self stabilizing homopolar induction bearing with integrated touch down
bearings has been developed for high-speed applications like flywheels, small
gas turbines and compact vacuum cleaners.

Stability is achieved without any control electronics thanks to stabilizing

eddy currents induced by permanent magnets. Eddy current losses are
reduced to a minimum using a homopolar design with ring magnets instead
of multipole or halbach arrays.

The bearing currents and forces are simulated using steady state 3D-FEM
analysis, which is enabled thanks to the implemented Minkowskij transform.
From these results an analytical model has been developed, and the results
are compared.

Efforts are made to develop a qualitative understanding of the bearing

physics. Results are converted into useful rotordynamic data that is easily
understood by machine engineers.

Finally some experiences from the first experimental test runs at 90.000 RPM
are discussed.

I


Acknowledgement
The work behind this dissertation was done as a joint project between the
Royal University of Technology, Stockholm, the Department of Electrical
Machines and Power Electronics and Magnetal AB (Publ), Uppsala.

The project was sponsored by the Swedish National Board for Industrial and
Technical Development, NUTEK, and Magnetal AB (Publ).

The author wants to acknowledge professor Chandur Sadarangani at the

Department of Electrical Machines and Power Electronics for valuable help
concerning the theoretical analysis of the models, and for his patience with
other urgent business matters that may have delayed this project.

Many thanks to Mats Leksell, my colleague and omnipresent roommate,

teacher in electrical machines, who helped me with the interpretation of the
induction bearing in terms of induction generator terminology.

Professor Hans-Peter Nee has been of great help in cross checking my

theories. He also helped me to bring understanding to some of the
complicated 3D effects, which I gladly, and without comment, will neglect in
this report.

Cheerful and encouraging comments were provided en masse by Peter

Bennich, who as a power quality researcher immediately saw the possibilities
with magnetically levitated flywheels.

I ought to compensate the laboratory staff of Magnetal with an additional

month of vacation for all their overtime. Marcus Granström has been of great
help in making the 3-dimensional CAD pictures of the bearings, and Per
Uselius was responsible for the mechanical engineering work on the
prototypes. Marcus also made the pictures of the futuristic machines in the
design study.

A new motor was developed for the high speed test spindle, and I especially
want to thank Louis Lefevre for his thorough analysis of that motor, and to
Dag Bergkvist at Magnetal who did a great job in trying to understand the
hand written winding diagrams from the author.

Stockholm December 2002

Torbjörn A. Lembke

II
 

Contents
 *(1(5$/,1752'8&7,21
1.1 HIGH SPEED DRIVES AND NEW BEARING REQUIREMENTS .................1
1.2 CONVERTING CONVENTIONAL MACHINES FOR HIGH SPEED .............3
 0HFKDQLFDOOLPLWDWLRQV 
 6ROXWLRQV 
1.3 MAGNETIC BEARINGS OPENS UP FOR NEW MACHINE DESIGNS .........5
1.4 OVERVIEW OF COMMERCIAL MAGNETIC BEARINGS ........................7
 7KH(DUQVKDZVWDELOLW\FULWHULRQ 
 &ODVVLFDODFWLYHPDJQHWLFEHDULQJV$0% 
 3HUPDQHQWPDJQHWEHDULQJV30%
 6XSHUFRQGXFWLQJEHDULQJV6&% 
1.5 MARKET NEED FOR LESS COMPLICATED MAGNETIC BEARINGS ......12
1.6 RESEARCH MOTIVATIONS AND OBJECTIVES ...................................13
 (/(&752'<1$0,&5(38/6,21±7+(.(<72
6,03/,&,7<" 
2.1 DRAWBACKS .................................................................................15
 35(9,286:25.21(''<&855(175('8&7,21
3.1 THE NULL FLUX SCHEME ...............................................................16
3.2 HOMOPOLAR CONCEPTS ................................................................19
 $[LDOIOX[FRQFHSW 
 5DGLDOIOX[FRQFHSW 
 +RPRSRODUEHDULQJVZLWKURWRUZLQGLQJV 
3.3 AXIAL FLUX GYROSCOPE STABILIZED BEARINGS ...........................24
 *(1(5$/21(''<&855(176$1'/266(6,1
+20232/$5(/(&752'<1$0,&%($5,1*6
4.1 CONCENTRIC OPERATION ..............................................................25
4.2 EXCENTRIC OPERATION. COMPARISON WITH INDUCTION
GENERATORS.................................................................................27
4.3 THE PRINCIPLE BEHIND ZERO LOSS BEARING OPERATION ..............31
 6HSDUDWLRQRIOLIWIRUFHDQGVWLIIQHVV 
4.4 PERPETUUM MOBILE –HOW CLOSE CAN WE COME?.......................33
 '(6,*12)7+(5$',$/)/8;+20232/$5,1'8&7,21
%($5,1*
5.1 GENERAL COMMENTS ON BEARING DESIGN ...................................34
5.2 BEARING DESCRIPTION AND DRAWINGS ........................................35
5.3 INNER ROTOR BEARING FOR SPINDLES AND TURBINES ...................36
 2XWHUURWRUEHDULQJIRUIO\ZKHHOV

III


 ,QWHUPHGLDWHURWRUEHDULQJIRUKROORZVKDIWVDQGSXPSV 

 %($5,1*$1$/<6,6
6.1 PURPOSE OF THE ANALYSIS ...........................................................40
6.2 MAGNETIC CIRCUIT .......................................................................41
 (GG\FXUUHQWVLQGXFHGGXHWRHFFHQWULFLW\
 (GG\FXUUHQWVLQGXFHGE\ODWHUDOPRWLRQ
 6WUD\HGG\FXUUHQWV
6.3 BEARING FORCES ..........................................................................55
6.4 LOSSES ..........................................................................................59
6.5 STIFFNESS .....................................................................................59
6.6 LOAD RANGE ................................................................................59
6.7 ROTORDYNAMICS ..........................................................................60
6.8 AUXILIARY BEARINGS ...................................................................63
6.9 COMPARISON WITH BALL BEARING STANDARDS............................63
 )(06,08/$7,216 
7.1 SIMULATION METHOD ...................................................................65
 6RIWZDUH
 3ODWIRUP
7.2 MODEL GEOMETRY OF THE STUDIED BEARINGS.............................67
7.3 SYMMETRY PLANES AND BOUNDARIES ..........................................68
7.4 MESH GENERATION .......................................................................69
 0HVKJHQHUDWLRQLQ0HJD 
 &RPPHQWVRQDXWRPDWLFPHVKJHQHUDWLRQ 
 6TXDUHRUWULDQJXODUHOHPHQWV" 
 6WUXFWXUHRIWKHPDQXDOO\JHQHUDWHGPHVK 
 ,QIOXHQFHRQPHVKVL]H 
7.5 SOURCES AND REGIONS .................................................................74
7.6 EDDY CURRENT DISTRIBUTION ......................................................75
 )LHOGGLIIXVLRQ
 ,PSRUWDQFHRIIOX[GHULYDWLYHV 
7.7 BEARING FORCES ..........................................................................79
7.8 BEARING LOSSES ...........................................................................86
7.9 PARAMETRIC OPTIMIZATION OF BEARING GEOMETRY ...................88
 6FDOLQJ 
 $GGLQJLQWHUPHGLDWHPDJQHWV 
 2SWLPL]DWLRQRIPDJQHWOHQJWK
 2SWLPL]DWLRQRIURWRUOHQJWK 
 2SWLPL]DWLRQRIPDJQHWZLGWK
 /HDNDJHIOX[UHGXFWLRQ 
 (;3(5,0(17$/9$/,'$7,212)7+(352727<3(6 
8.1 THE BEARING PROTOTYPES .........................................................102
8.2 TEST RUNS...................................................................................104
 &21&/86,216$1')8785(:25. 

IV
 

5()(5(1&(6 

$33(1',; 
SIMPLE HANDS ON RULES FOR DESIGNING INDUCTION BEARINGS .............112
$33(1',; 
MODEL GEOMETRIES ................................................................................114
$33(1',; 
PATENTS ..................................................................................................115

V


List of symbols
Symbol Unit Description

% 7 Magnetic flux density

%U 7 Remanent flux density
GL P Inner diameter of inner magnet
& 1VP Damping coefficient
'L P Outer diameter of inner magnet
GU P Inner diameter of conducting rotor
'U P Outer diameter of conducting rotor
GR P Inner diameter of outer magnet
'R P Outer diameter of outer magnet
I +] Frequency
 )[ 1 Force in x direction
) \ 1 Force in y direction
)] 1 Force in z direction
)U 1 Force caused by displacement ∆r
) Y 1 Force caused by in plane motion v
)G 1 Force caused by external damper
 JQ P Nominal airgap
. 1P Bearing stiffness
N\ 1P Bearing stiffness in y direction
OP P Axial length of magnet
OZ P Axial length of iron pole shoes at ends
Or  P Length of conducting bearing rotor
 P NJ Mass of rotor 
 P $P Magnetization
0[ 1P Moment about x axis
0\ 1P Moment about y axis
0] 1P Moment about z axis, brake moment
S  Number of poles in heteropolar bearing
ZP P Width of magnet cross section.
[ P Coordinate axis perpendicular to displacement
\ P Coordinate axis parallell to displacement
] P Coordinate axis about which bearing is
symmetric
 α GHJ Angle of force resultant
ρ ΩP Resistivity of rotor
ρ P Radial polar coordinate

VI
 

θ UDG Force angle, defined as the complimentary angle

between the force caused by a certain
displacement.
σ ΩP Conductivity of rotor
τ P Pole width
ω V Rotational speed

VII


VIII
 *HQHUDOLQWURGXFWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

1 General introduction
 +LJKVSHHGGULYHVDQGQHZEHDULQJUHTXLUHPHQWV
The benefits of using modern high-speed technology instead of using
conventional machines are becoming apparent to an ever-increasing number
of engineers. Higher power density and better efficiency are advantages
usually sought for, thus leading to smaller machines with lower power
consumption. Hand held screwdrivers and mobile power generators are
applications where these advantages become obvious, and the number of
applications is likely to increase in the future.

What is the definition of a high-speed machine? Some years ago the answer
would definitely have been that there is a sharp line between motors running
at network frequency and motors running faster. However, with the widely
spread use of inverters the limit is now moving upwards. During the past ten
years turbomolecular pumps operating at 90,000 RPM have been mass-
produced showing that the technology is mature. These pumps are often
referred to as high-speed turbo pumps.

Recently the term very high speed machines was launched referring to the
recent research at MIT on small 0.1 kW turbogenerators operating at between
1 and 2 million RPM.

Probably the fastest machine ever made was a motor made for splitting steel
balls. It levitated the balls in a magnetic field and forced them to rotate at 18
million RPM when they finally burst.

An engineer designing a high-speed machine will have to tackle other types

of problems than usually treated by courses in electrical engineering.
Questions like how to reduce air drag losses and how to eliminate eddy
current losses or how to treat problems concerning rotor dynamics and
vibration control are likely to occur.

A special type of problem that many engineers regard to be the most difficult
one, is the choice of bearings. At high speed the lifetime of a ball bearing is
very limited. This may not matter for some applications as is the case for a
hand held screwdriver that only operates a few seconds at a time, but for
other applications the lifetime is crucial.

In some environments the noise level determines the choice of bearings. If

the rotor is not perfectly balanced, or maybe it creeps with time and gets

1
*HQHUDOLQWURGXFWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

unbalanced after a while, a bearing with high stiffness will transfer the
vibrations to the housing and should thus be avoided.

In other cases the bearing must be lubrication free, for instance in

hydrocarbon free environments, or where the customer demands maintenance
free operation.

Bearings like ceramic ball bearings, air bearings or fluid bearings are often
used, all with their special limitations. It is not surprising that a new type of
bearing, WKHPDJQHWLFEHDULQJ, is becoming very popular in designing high-
speed machines.

A magnetic bearing is contact free and thus has very long lifetime. It is
lubrication free and thus maintenance free. It has low stiffness and thus does
not transmit vibrations to the housing. It is quiet. And it has very low losses,
even at very high speed. If it would not be for the price this bearing would
surely be the ultimate choice.

This report will deal with the possibility to make a low-price magnetic
bearing that still offers most of the advantages you expect from this emerging
and promising new bearing technology. The bearing will be referred to as the
radial flux homopolar induction bearing, fig. 1.


)LJ+RPRSRODULQGXFWLRQEHDULQJFRPSDUHGWRDVSKHULFDOEDOOEHDULQJ

2
 &RQYHUWLQJFRQYHQWLRQDOPDFKLQHVIRUKLJKVSHHG
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 &RQYHUWLQJ FRQYHQWLRQDO PDFKLQHV IRU KLJK

VSHHG

1.2.1 Mechanical limitations

When inverters are used to increase the speed of a conventional machine,
without changing the machine design, the bearing lifetime is reduced. Also
vibrational problems are likely to occur. At very high speed centrifugal
forces need to be taken into careful consideration. Below, we will start to
focus on bearing lifetime and vibrational modes, and in the next subsection
we will also add some aspects on centrifugal strength.

In a ball bearing the speed of the balls and their centrifugal forces increases
when inverters are used to raise the speed. To a certain limit using low
density ceramic balls instead of steel balls can reduce the centrifugal force,
but this does not help very much since the force increases linearly with
weight and quadratically with speed. So the best way to increase the lifetime
is often to reduce the diameter. If the load is too high for a smaller bearing,
two bearings can be used in line. In this case a so-called paired couple of
bearings is often used to reduce the bearing axial gap. The important thing is
to reduce the diameter and thus the peripheral speed of the bearing.

However, this method is contradictory to the desire to maintain the torque

while increasing the speed, since the same torque has to be transmitted
through a thinner shaft, which can lead to torsional vibrations and fatigue
stress. Thus the torque during normal operation in such applications has to be
reduced in order to enable the use of a thinner shaft.

Reducing the diameter has another disadvantage: it reduces the frequency of

the first bending mode, which makes high speed operation more demanding
and increases the need for appropriate damping. Fig. 2a.

3
*HQHUDOLQWURGXFWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

1.2.2 Solutions
The solution is to use large diameter bearings that still offer long lifetime.
This makes magnetic bearings the ideal choice. Fig. 2b. Today, magnetic
bearings are used in a variety of applications, so the main question is seldom
whether or not magnetic bearings can do the job. The question is what kind
of magnetic bearing, or bearing combination, that should be used to fulfill the
mechanical specifications within given economical restraints. This sometimes
results in interesting bearing combinations. Often used are combinations of
passive and active magnetic bearings. Another popular hybrid is the
combination of passive magnetic bearings and ball bearings. Finally air
bearings and passive magnetic bearings are sometimes used together. The
reader who is not familiar with these types of bearings is recommended the
chapter 1.4 “Overview of commercial magnetic bearings.”

With the introduction of cost effective induction bearings the number of

possible bearing combinations increase. The induction bearing is of particular
interest at peripheral speeds over 200 m/s. At this speed sheet steel rotors
used in active magnetic bearings can hardly withstand the centrifugal forces,
while an induction bearing with an aluminum cylinder can easily operate at
such speeds.


)LJ&RPSUHVVRUZLWKDEDOOEHDULQJVDQGEPDJQHWLFEHDULQJV

Thanks to the simplicity of the induction bearings they can open up for new
designs since they neither require rotating magnets nor bearing rotors of sheet
steel. Chapter 1.3 will show some examples of this type of bearings.

4
 0DJQHWLF%HDULQJVRSHQVXSIRUQHZPDFKLQHGHVLJQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 0DJQHWLF %HDULQJV RSHQV XS IRU QHZ PDFKLQH

GHVLJQV
If magnetic bearings are considered early when developing a new machine,
completely new designs are possible. Often there is no need for a shaft at all.
The bearings and the motor can then be integrated into the load, thus
reducing the size and the mechanical complexity of the whole machine. This
concept is especially useful for gasturbines, centrifuges, pumps, flywheels,
gyroscopes and other direct driven applications.

Upper axial
Eddy current
bearing
damper

Motor/ Radial
generator induction
bearings


)LJ+LJKVSHHGIO\ZKHHOIRUHQHUJ\VWRUDJH


A flywheel without shaft as in fig. 3 has obvious advantages over
conventional flywheel designs. The drive and the bearings can be integrated
in the design without the need for a long thin shaft. In this particular design
two radial homopolar induction bearings and one passive axial bearing of
permanent magnet type are used. The drive is an outer rotor permanent
magnet airgap winding motor developed by Magnetal.

The shaft is often the weak point in conventional flywheels, due to bending
problems at high speed and due to the very large torque when the machine
operates as a generator. Especially flywheels used for power quality
applications, which have to deliver very high power during short periods,
suffer from these torque peaks.

5
*HQHUDOLQWURGXFWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB


)LJ*DVH[SDQGHUZLWKLQWHJUDWHGGULYHXQLW

Turbines is another group of applications where magnetic bearings offer

interesting design possibilities. Fig. 4 shows an interesting picture of a gas
expander from a design study made by Magnetal. The bearings and the drive
unit are integrated in the impeller.

Avoidance of a flexible shaft with additional damping arrangements, and

elimination of the complex oil jet injection lubrication systems increases the
reliability of the total system as the number of moving parts are reduced.
Also the price of the system as a whole will probably be lower than for
conventional turbine designs. If low maintenance requirements are crucial,
the magnetic bearing is the natural choice.

Active magnetic bearings will most likely do the job for most applications,
but for the designs in fig. 3 and 4 it is important to get rid of the emergency
ball bearings. In these cases induction bearings with integrated touch down
bearings are better suited in order not to have to decrease the diameter or to
increase the length of the shaft.

6
 2YHUYLHZRIFRPPHUFLDO0DJQHWLF%HDULQJV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 2YHUYLHZRIFRPPHUFLDO0DJQHWLF%HDULQJV
A magnetic bearing is a contact free bearing wherein the load is carried by
magnetic forces. Furthermore the magnetic field is generated in such a way
that it provides necessary stiffness and damping for the rotor to be levitated
safely during operation.

Magnetic bearings have been available on the market for about 30 years.
Usually they are divided into two categories, often referred to as either active
or passive depending on whether an electronic control system is being used
or not.

This nomenclature has been around for quite a while even though it is rather
unfortunate and sometimes gives rise to misunderstandings.

It would certainly be more clarifying to classify magnetic bearings according

to the physical cause of the force, as H. Bleuler has done [A]. He divides the
bearings in two groups, reluctance force bearings and electrodynamic
bearings. Then each group is divided into five respectively four categories
depending on how the magnetic force is generated. For a brief presentation of
each category the reader is recommended the book Active Magnetic Bearings
by G. Schweitzer [B].

Some bearing categories are of only limited academic interest, while others
have found many useful applications. Based on Bleulers classification some
of the most technically and commercially important bearing types will be
presented below. But first we need to take a closer look at the general
stability criterion for magnetic bearings.

1.4.1 The Earnshaw stability criterion

As early as 1842 Earnshaw [C] was able to show that it is impossible for an
object to be suspended in stable equilibrium purely by means of magnetic or
electrostatic forces.

Let V be the scalar magnetic potential at a point, proportional to the scalar

product of % at that point and the constant magnetization P for a single
magnetic pole on the suspended object. Then according to the Laplace
equation (1) the sum of the second derivatives is zero,

G 29 G 29 G 29
+ + =0. (1)
G[ 2 G\ 2 G] 2

7
*HQHUDOLQWURGXFWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

In order for a magnetic pole to experience no magnetic force and have no

tendency to move, it is necessary that

G9 G9 G9
= = = 0. (2)
G[ G\ G]

This is the condition for a maximum or a minimum in the potential V. Only

at a potential minimum is the equilibrium stable. The condition for a potential
minimum is that its second derivative is positive. At best, two of the second
derivatives can be made to be positive, then, according to the Laplace
equation (1), the third has to be negative. In that direction the equilibrium
will be unstable. The theorem has been derived for a single pole, but it should
apply to a magnet as a whole.

In order to achieve stability in all directions it is necessary to add a dynamic

term, which can be either an electromagnetic or a mechanical one. Adding an
electromagnetic term will exchange the Laplace equation for the Poisson
equation, which is used in active magnetic bearings and electrodynamic
bearings. When a mechanical force derivative is added it can be done
directly into equation (1). For the bearing to be of the non-contact type the
preferred mechanical forces are either gyroscopic or fluid dynamic.

Gravitation is for instance not possible to utilize as a means of stabilizing the

bearing, since it has only negligable derivatives. Say that we want to
compensate the vertical magnetic force )\ with a gravitational force –PJ
Then the total potential energy 9WRW should be used instead of the magnetic
potential V, the latter being denoted Vmag below to avoid confusion.
Unfortunately this does not change anything in the Laplace equation (1) since
the second term remains unchanged as shown below:

G 29WRW G2 G2
G\ 2
=
G\ 2 ∫ )\,WRW G\ = G\ 2 ∫
( ) \ , PDJ − PJ )G\ =
(3)
G)\ , PDJ G G)\ , PDJ G 29 PDJ
= − (PJ ) = =
G\ G\ G\ G\ 2

Thus, gravity does not improve the stability of a levitating magnet.

What is said above can be applied to other constant forces as well, like air
drag in a windmill. They cannot be fully levitated by permanent magnets
alone. However, other air forces like the wedge effect in an air bearing can be
used for stabilisation since that force has a space derivative defined by the
airgap.

8
 2YHUYLHZRIFRPPHUFLDO0DJQHWLF%HDULQJV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

1.4.2 Classical active magnetic bearings, AMB

An active magnetic bearing is a typical example of an advanced mechatronics
product. All AMB:s on the market today utilize attracting reluctance forces
generated by strong stator electromagnets acting on a ferromagnetic rotor.
The magnetic equilibrium reached in this way is inherently unstable. Thus an
analog or digital controller is used to stabilize the bearing. Contactless
sensors are used to measure the rotor position. Any deviation from the
desired position results in a control signal that via the controller and the
amplifiers changes the current in the electromagnet so that forces are
generated to pull the rotor back so that stability is maintained.

Currently only attractive forces are used, but theoretically also repulsive
forces could be used for actively controlled levitation. This is the reason why
active bearings are sometimes referred to as attracting bearings, while passive
bearings are often referred to as repulsive.

The advantages with active bearings over passive bearings are obvious as
they offer the possibility to change and adapt the controller algorithms
according to machine dynamics. The disadvantages are equally clear and can
be spelled in four letters: cost. Also reliability is a problem as the controller is
dependent on high power quality and thus need energy back up.

Today active magnetic bearings are found in a variety of applications like

compressors, gas turbines, motors, flywheels, gyroscopes, fans and machine
tool spindles.

1.4.3 Permanent magnet bearings, PMB

One of the first successful commercial products where magnetic bearings
offered true added value to the customer was the turbomolecular vacuum
pump. With lubrication free magnetic bearings it was possible to manufacture
hydrocarbon free pumps -a great achievement at that time.

The first mass produced pump of this kind had one passive magnetic bearing
on the high vacuum side and one ball bearing on the exhaust side. The
magnetic bearing consisted of concentrically arranged repulsive ring
magnets, fig. 5. To the left a bearing with a single magnet pair, and to the
right to oppositely directed pairs are used. The latter configuration gives
almost 3 times the stiffness compared to the former one. An optimization of
PMB geometry was recently done by Lang and Fremerey8, who showed that
many thin magnets should be used instead of a few thick ones.

9
*HQHUDOLQWURGXFWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

Rotor ring magnets

Stator ring magnets



)LJ3DVVLYHPDJQHWLFUDGLDOEHDULQJV

For low speed applications it is convenient to let the inner rings be mounted
on the rotor, while for high speed applications it is better to let the outer rings
rotate due to the centrifugal forces, in which case these forces are used more
advantageously in keeping the rotating magnets in place.

When the term passive magnetic bearing is mentioned in this report as well
as in most literature this is the magnet arrangement usually referred to.

Attractive magnetic forces from permanent magnets could have been used as
well, but as the geometry for such bearings is slightly more complicated they
are seldom used.

All passive bearings of these kinds have one thing in common. According to
the Earnshaw theorem they are unstable in at least one direction. For the
pump mentioned above the repulsive bearing is strongly unstable in the axial
direction. This is compensated for by the use of one ball bearing that prevents
the bearing from moving axially.

The next progress was made when the axial stability from the ball bearing
could be exchanged for an active magnetic axial bearing. In this case two
repulsive radial bearings were used. This concept was originally developed
by S2M in Vernon for flywheels for gyroscopic stabilization of satellites.
Later, Dr. Fremerey in Julich modified this concept for use in the
turbomolecular vacuum pumps manufactured by Leybold.

We have now mentioned that active bearings and ball bearings can stabilize
passive bearings. Other possibilities are to combine passive bearings with
electrodynamic or superconducting bearings, or to stabilize them with
gyroscopic effects like in the well known toy, the "Levitron", se fig. 14.

10
 2YHUYLHZRIFRPPHUFLDO0DJQHWLF%HDULQJV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

1.4.4 Superconducting bearings, SCB

A superconductor has by definition two properties that makes it ideal for
magnetic bearings. The first one is the Meissner effect which makes the
superconductor act like a perfect diamagnet. A permanent magnet placed
above a superconducting surface will "see" a mirror image under the surface
polarized so that the force is repulsive. This levitation is stable in the vertical
direction, and indifferent in the horizontal direction. If the real magnet moves
the image moves too, thus maintaining the repulsive force vertical.

v
v=0 B Real magnet

F Superconducting surface

Image magnet

)LJ5HSXOVLYHIRUFHGXHWRWKHPHLVVQHUHIIHFW

The modern high temperature superconductors have a built in defect that at a

first glance makes them even more interesting for magnetic bearings. If the
magnet is placed close to the surface before the superconductor has been
cooled down to its superconducting state, not all the flux is expelled when the
temperature is decreased again. This method is called field quenching. Some
of the field is "frozen" into the superconductor due to an effect called flux
pinning. This frozen field stabilizes the magnet in both radial and axial
direction. If the surface is turned upside down the magnet does not fall down.

The author, at that time employed by SKF, evaluated this effect in 1990 for
use in magnetic bearings, but it was found that the long time stability of such
bearings was limited. The author predicted that when used in a rotating
machine exposed to vibrations, then according to the hysteresis loop caused
by flux pinning the bearing would slowly move in the direction of the static
load and eventually touch down. Later research on superconducting bearings
at Argonne national lab and many other places have proved this to be true.
This is regarded to be a major drawback for this bearing technology.

The second effect defining a superconductor is the zero resistance. This has
been used to make strong electromagnets for magnetically levitated test
trains. In this case the bearing principle is not the superconducting one, but
the electrodynamic one. The superconductors are just used to replace the
permanent magnets that would otherwise have been used.

11
*HQHUDOLQWURGXFWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

It is possible to use the superconductor as the conducting member in an

electrodynamic bearing. However, due to large hysteresis losses this may be
found not to work properly. As is nowadays well known, the AC losses in a
superconductor are quite large.

 0DUNHW QHHG IRU OHVV FRPSOLFDWHG PDJQHWLF

EHDULQJV

There is an increasing demand for small uncomplicated stand alone bearings

for various mass produced applications. It can be applications where noise
reduction is important, like computer fans and hard disc drives.

Oil free bearings are important in chemical, biomedicine and biotech

applications. Hydrocarbon free environment is necessary in analytical
instruments, and in the large electronics manufacturing area.

Maintenance free operation is important in unmanned stations for telecom,

television and space industry. Space industry can still afford active magnetic
bearings, but the highly competing telecom market need to reduce investment
costs to a minimum.

Low energy loss is important in all kinds of energy generation and energy
storage applications. It is also important for applications where energy is
limited like in battery-powered electronics and vehicles.

Despite the market need, there are no magnetic bearings available on the
market that does not require either an electronic control unit or cryogenic
cooling equipment. Thus there are no bearings available for the huge markets
that small and mass-produced products represent.

Current research and development of stand alone induction bearings is an

attempt to fill this giant gap between on one side market and environmental
need and technological achievements on the other.


12
 5HVHDUFKPRWLYDWLRQVDQGREMHFWLYHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 5HVHDUFKPRWLYDWLRQVDQGREMHFWLYHV
It seems clear that high-speed technologies will play an important role in the
not too far future. It might even be apt to talk about a technology shift where
heavy, noisy and dirty industrial machines are exchanged for small high-
speed clean and silent ones.

Distributed energy production is an example where large amounts of small

gasturbine driven generators, maybe in connection with flywheels for peak
power, can prove to be the ideal solution for electrifying large suburban
areas. Such a vision is not possible to realize without maintenance free high-
speed bearings in both turbines and flywheels. Also rapidly emerging
technologies related to Internet and telecom require better power quality than
the power grid alone is able to supply.

Though active magnetic bearings, AMB´s, may seem to be the obvious

choice from a technical point of view, they do not always convince the
purchaser. Why should he exchange two conventional ball bearings for three
other ball bearings and three magnetic bearings? And why should he pay for
emergency bearings at all when they are never supposed to be used? Other
questions he may ask are: Will AMB:s ever be cheap in mass production?
What about training programs for the staff? AMB:s do not exactly behave
like conventional bearings.

When this project started it was felt that if there will ever be a great
breakthrough for high speed applications in general and magnetic bearings in
particular, someone needs to look at these questions seriously.

In 1996 the Swedish National Board for Industrial and Technical

Development, NUTEK, and Magnetal AB (Publ) agreed on funding the
current research project about a new type of induction bearing based on the
homopolar design.

The objectives of this research can be condensed into the following lines:

• To simulate the magnetic properties of the homopolar induction bearing

using 3-dimensional finite elements.
• To develop a qualitative understanding of the bearing physics.
• To convert the results into useful rotordynamic data which can readily be
used by machine engineers.

13
(OHFWURG\QDPLFUHSXOVLRQ±WKHNH\WRVLPSOLFLW\"
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

2 Electrodynamic repulsion –
the key to simplicity?
Electrodynamic bearings is a large group of magnetic bearings, all of them
based on the electrodynamic repulsion principle. They offer many similarities
with induction generators, which is the reason they are sometimes called
induction bearings. Eddy current bearings is a term often used in patent
applications, also referring to this group of bearings.

Like superconducting bearings, induction bearings can provide stable

levitation without control electronics. Unlike superconductors they do not
need cooling, which make them very attractive from a commercial point of
view. Like fluid film bearings and many other bearings the lift force is speed
dependent.

The electrodynamic principle is based on Lenz´ law. According to this law a

change of magnetic flux through a conductor will induce a voltage in the
conductor in such a way that a current is produced that tries to maintain the
magnetic flux at the original level. In reality this means that any changing
electromagnetic flux will be reflected on a conducting surface, thus giving
rise to repulsive forces. The maximum force is generated at infinitely high
speed. This force can be calculated as if there is a mirror image of the real
magnet on the opposite side of the surface. This is very much like the
superconducting effect described earlier, and the bearings are equally strong,
at least for high frequencies. At low frequency an electromagnetic bearing
has very low lift force. Thus at low speed induction bearings will touch down
on landing bearings.


v
v=0 B Real magnet

F=0 F
. x
Conducting surface I+ I-

Image magnet



)LJ7KHHOHFWURG\QDPLFOHYLWDWLRQSULQFLSOH

Fig. 7 illustrates the electrodynamic levitation principle applied to a linear
bearing where a moving magnet levitates above a conducting plate. The plate

14
 'UDZEDFNV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

could as well be a set of short circuit coils. At zero speed there is no change
of flux and thus no eddy currents are induced and no force arise. At high
speed eddy currents I+ and I- are induced and the mirror effect develops.
However, the mirror is incomplete and phase shifted due to the finite speed
and the resistance in the conducting plate, fig. 8 right. The phase shift
changes the angle of the resultant force F so that there is also a pushing brake
force component. The force F acts between the magnet and the induced
currents, but can also be calculated as the force between the real magnet and
its image, which was done by the author4 in 1990. It may seem a little bit
confusing that the mirror comes ahead of the real magnet, but this is a natural
result from the fact that the eddy currents are induced from the change of the
flux from the real magnet, which always comes before the flux itself. The
brake force is well known from linear induction motor theory where it is
reversed and used for traction.

Instead of moving permanent magnets an alternating magnetic field can be

generated by coils connected in a resonant circuit. However, the ohmic losses
in such a bearing tends to be rather high, so such bearings are mostly
investigated for very small or very light applications. The large ohmic losses
is the reason for why permanent magnets are usually used instead of
electromagnets.

If extremely large forces are desired, the permanent magnets can be

exchanged for superconducting coils carrying DC -current. This is the case
for some linear electrodynamic bearings that have been proposed for a
variety of large magnetically levitated vehicles like trains and rocket
launchers. These devices are often referred to as maglevs. In these cases
strong superconducting magnets are needed to produce the desired lift force.
Some of these applications have already been built and tested and are likely
to be of some importance for future transportation.

 'UDZEDFNV
Trying to apply electrodynamic bearings for small and medium size rotating
machines is not as easy as for large ones. Scaling properties makes it difficult
to reduce the losses and an attempt is likely to result in an induction heater
rather than in a functional bearing.

Many attempts to decrease the eddy current losses have been done in the past,
and a review will be given in chapter 3. However, though many inventors did
great work, none of their bearings are available on the market today, the
remaining losses being the main issue.

15
3UHYLRXVZRUNRQHGG\FXUUHQWUHGXFWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

3 Previous work on eddy

current reduction
 7KHQXOOIOX[VFKHPH
Different authors, like Powell and Danby, have proposed some means of
reducing eddy current losses in linear electrodynamic bearings for
magnetically levitated trains, maglevs. They introduced what they called the
“null flux scheme”, fig 8b, which means that the magnets can be arranged in
such a way that they do not produce losses until they are needed. Compare
fig. 7 and fig. 8a. In fig. 8 an additional lower magnet is added which induces
a second eddy current on the lower surface of the conducting plate. The upper
and the lower currents do not interfere, since the skin depth prevents them
from penetrating into each other. However, in fig. 8b the conducting plate is
thinner than one skin depth so that the eddy currents almost completely
eliminate each other. This is true as long as the plate is centered between the
magnets in the “null flux region” where there is no flux perpendicular to the
plate. When any deviation occurs, the field is no longer zero and repulsive
eddy currents are induced.

v
B Upper magnet v
Upper magnet

I- I+
. x
x . I ≈0
I+ I-
v
v Lower magnet
Lower magnet

)LJDE7KHQXOOIOX[VFKHPH

Sacerdoti et al3 utilize this effect in a thrust bearing which he combined with
passive radial bearings and reported losses that were “not too high”.

In 1990 the author4 demonstrated a prototype with eddy current bearings

where the linear null flux scheme was applied to axial as well as radial
bearings. The rotor weight was 1 kg and the speed was 12 000 RPM. In order
to further minimize the losses a minimum of magnets was used, which
resulted in a low stiffness. Despite the low stiffness it was possible to

16
 7KHQXOOIOX[VFKHPH
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

demonstrate full magnetic levitation, using only permanent magnets and

rotating aluminum discs and tubes. The weight of the rotor was unloaded
from the bearings by a separate set of permanent magnets. Thus the lowest
possible speed was not determined by the lift force, since the separate lift
magnets produced a constant lift force, but by the rotordynamic stability. At 8
000 RPM the rotor centered and the maximum speed tested was 14 000 RPM.

It was also possible to safely run through critical speeds thanks to integrated
plain bearings/dampers. The concept was promising, but the losses were still
far too high.

Between 1991 and 1995 much work was done by the author to increase the
stiffness and to reduce the losses. During this period the basic design that was
used for both thrust bearings and journal bearings was the heteropolar eddy
current bearing with permanent magnets. Due to the similarities with an
induction generator with large slip the bearing was called Magnetic Induction
Bearing, MIB, or simply induction bearing.

The goal during this period was to make a thrust bearing that produced a
stiffness of 1 N/mm for each Watt of power dissipation at a speed of 100 m/s
measured as the speed over the magnets. When this goal was reached the
research project was ended, although it seemed possible to reduce the losses
even further.

17
3UHYLRXVZRUNRQHGG\FXUUHQWUHGXFWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

)LJ+HWHURSRODULQGXFWLRQEHDULQJVGHVLJQHGE\WKHDXWKRUDW6.)LQ

In 1998 R.F.Post6 presented an interesting solution to the eddy current

problem in heteropolar radial bearings by simply using a wire wound rotor
with the coils connected to minimize the undesired currents. In fig. 10 a
prototype of Post´s bearing is shown, modified and built by the group of
Passive Magnetic Bearings in Lausanne. For axial stabilization passive axial
repulsive bearings are suggested.

Instead of reducing the radial flux as Sacerdoti, Powell and Danby and the
author did, he series connected the coils in such a way that the induced
voltages eliminated each other.

18
 +RPRSRODUFRQFHSWV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

)LJ$SURWRW\SHIURP/DXVDQQHZKHUH5LFKDUG3RVWVEHDULQJLVHYDOXDWHG


All these concepts, including the latter solution from Post, can all be
described as multipole or heteropolar bearings. They all work according to
the principle of eddy current reduction by letting either the induced voltages
or the penetrating fluxes counteract each other. The result is the same.

A drawback with the null flux scheme presented by Powell and Danby is that
the null flux region is infinitely thin. Thus the conducting part of the rotor,
whether it is a bulk material or made up of coils, also needs to be infinitely
thin in order not to produce any losses. This in turn requires that the speed is
infinitely high; otherwise the rotor can’t repel the flux lines.

A practical design needs a certain material thickness. What was needed was a
bearing concept that allows a strong flux to penetrate the conductor without
inducing unnecessary eddy currents. This leads to solutions involving
homopolar magnetic flux.

 +RPRSRODUFRQFHSWV
Recent research in Japan, Schwitzerland and USA has shown that it is
possible to get around some of the problems above by using a homopolar
magnetic field instead of a heteropolar field.

19
3UHYLRXVZRUNRQHGG\FXUUHQWUHGXFWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

In a bearing with a homopolar field there are no induced voltages that need to
be eliminated like in the multipole bearings above. Filatov9 introduced the
term “Null-E magnetic bearings” for this bearing concept, to distinguish it
from the previously described “Null flux bearings”. The circular design
avoids the induction of any unnecessary eddy currents, as long as the rotor is
centered in the bearing stator. The result is the same, but the homopolar
concept is much simpler as no winding is required to avoid all the
unnecessary eddy currents. Homopolar bearings, as we will denote them, can
be of either of the axial flux or radial flux type.

3.2.1 Axial flux concept

A simple axial flux homopolar bearing can look like the left bearing in fig.
11. The advantage of an axial type design is that the magnetic flux path is
very effective with a relatively small airgap. Eddy currents are induced in the
bearing disc, which rotates in the airgap. The disadvantage with this design is
that the eddy current path is rather ineffective thus increasing the total
resistance of the electric circuit. Another disadvantage is that the disc can
only utilize about half of the flux, at a given instance of time. The remaining
flux is used to build up an area with a strong flux gradient in which the rotor
can move when it is not centered.




+RPRSRODULQGXFWLRQEHDULQJV

$[LDOIOX[W\SH 5DGLDOIOX[W\SH

B ω S
N N

N
S
S

Iron yoke Conducting disc Conducting cylinder Iron yokes

)LJ

20
 +RPRSRODUFRQFHSWV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

3.2.2 Radial flux concept

The current research by the author on radial flux homopolar bearings is an
attempt to overcome some of the abovementioned problems with axial flux
bearings. A comparison between the axial flux and the radial flux concepts is
made in fig. 11.

The radial flux concept differs from the axial flux bearing in that the rotor has
no disc, but only a conducting cylinder into which the flux penetrates in
radial direction, fig. 11 right. If a typical flux path is considered, the effective
airgap is larger than for the axial flux bearing. However, the active length of
the eddy current path is relatively shorter than in the axial flux bearing, thus
resulting in a more effective concept. From fig. 11 it is also clear that the ra-
dial flux concept is more compact and requires less space than the axial one.

)LJ6ZLVVURWRUZLWKDLUWXUELQHDQGKRPRSRODUEHDULQJV

During the Seventh International Conference on Magnetic Bearings in Zurich

in August 2000 it was found that parallel to our research the homopolar
concept is also being studied in Schwitzerland by the Group of Passive
Magnetic Bearings in Lausanne.

21
3UHYLRXVZRUNRQHGG\FXUUHQWUHGXFWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

During the conference Hannes Bleuler and Jan Sandtner from the group
presented some results from measurements at low speed up to 9 000 RPM,
using an air turbine driving a shaft with two radial flux homopolar bearings,
fig. 12. According to our research, a speed of 9 000 RPM is a little bit too
low to be able to observe the interesting high speed phenomena that is
studied in this report. The group intended to increase the speed in future
research, and it will be very interesting to compare their results with ours for
validation.

3.2.3 Homopolar bearings with rotor windings

Instead of bulk rotor disks and cylinders it is possible to use rotating
shortcircuited windings instead. The advantage is that the induced eddy
currents can be lead into desired current paths, and that stray eddy currents
due to badly oriented magnets thus will have a much less negative effect on
losses. With coils it is also possible to series connect an electric circuit in
order to improve certain bearing characteristics.

Filatov and Maslen9 have shown that a simple circuit like a series inductance
has a positive influence on low speed stability for axial type homopolar
bearings. This is due to the increased inductance, which in turn decreases the
force angle α.

Filatov made a bearing prototype, fig 13, which was able to levitate at speeds
from 21 - 40 Hz. The low speed limit was set by the damping properties, and
the high speed limit by the material strength. Damping turned out to be a
cruicial point, so he had to add additional damping by connecting stator
mounted damping coils to an external electrical circuit.

The obvious disadvantage with rotating coils and rotating circuits is that they
are not well suited for very high speed applications due to the centrifugal
load on each component. In this aspect bulk conductors are much better. For
this reason Filatov and Maslen had to optimize their bearings for low speed
operation, though with better rotor materials their bearings could certainly
have been used at higher speed too. At higher speed he would likely not have
needed the additional damping electronics, since damping requirements
decreases with increasing speed.

Filatovs and Maslens work is important in more than one way. Though
nowadays quite a few reports on passive levitation are published, very few of
them contain any analytical explanations to the bearing behaviours. Filatovs
and Maslens work includes several analytical approaches that forms a solid
foundation for future work. Thus in this report we will try to adopt their
nomenclature to our bearing analysis, though the focus in this report lies not
in mathematical but in finite element analysis of bearing properties.

22
 +RPRSRODUFRQFHSWV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB



)LJDE)LODWRYVKRPRSRODUEHDULQJZLWKURWRUZLQGLQJV



'HVFULSWLRQ

1 Multi turn loops 11 Series inductor

6 Damping coils 12 Magnetic yoke
7 Outer stator stabilizing magnet 15 Rotor magnet
8 Inner stator stabilizing magnet 16 Lift magnet
9 Steel disc 17 Tilt protection magnets

23
3UHYLRXVZRUNRQHGG\FXUUHQWUHGXFWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 $[LDOIOX[J\URVFRSHVWDELOL]HGEHDULQJV
Finally a few words will be said about another homopolar bearing, which is
spread around the world in a toy called the “Levitron”.

The levitron, fig. 15, consists of two concentrically, but axially displaced,
permanent ring magnets. The larger one is mounted in a fundament and the
smaller one is rotating mounted to the spinning top.

There are no eddy currents involved in stabilizing the top. Instead gyroscopic
forces and gravity stabilizes the bearing. However, stability is only achieved
within a very narrow speed range. At low speed when gyroscopic forces are
negligible, the Earnshaw stability criterion is applicable and the top tips
around and is attracted to the other ring magnet.

At high speed the rotor is stiff compared to the available magnetic forces, so
it can´t tip, and thus there is no possible cross coupling that can stabilize the
bearing. The problem can be said to be two dimensional, and the two
dimensional form of the theorem is still applicable.

The speed range depends on the size of the magnets, but lies for the most
common size between 20 and 30 rps, revolutions per second. This limits the
use of the bearing, and until now the invention has not found any other use
than in this fascinating toy.

)LJ7KH/HYLWURQ

24
 &RQFHQWULFRSHUDWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

4 General on eddy currents and

losses in Homopolar
Electrodynamic Bearings
In chapter 3 an overview of different ways of dealing with the problem of
eddy current losses in electrodynamic bearings was given. It was found that
recent research on different homopolar designs represents a technological
breakthrough for this kind of bearings. By introducing a rotationally
symmetric magnetic flux, similar to that used in homopolar electric motors,
these inventors have all shown that it is possible to reduce the induced eddy
currents to a minimum, virtually to zero.

In this chapter we shall focus on general principles that will help the reader to
understand how heavy rotors can levitate at high speed, almost without the
induction of any eddy currents. We will be able to show that eddy currents
can be divided into two groups; necessary and unnecessary ones. With the
homopolar bearing we are able to exclude almost all currents of the second
category.

Expressed in a more scientific, almost philosophic way, the subject of this

chapter is “to explain why bearing stability, due to the Earnshaw stability
criterion, requires the SRVVLELOLW\ to induce eddy currents, and yet, does not
necessarily need them to be induced”.

Important keys to understand is:

• why eddy currents are neither required, nor possible in concentric
operation
• how eddy currents are induced in excentric bearing operation
• how to separate bearing stiffness from bearing lift force with
permanent magnets

 &RQFHQWULFRSHUDWLRQ
A conducting body of arbitrary shape, rotating in a rotationally symmetric
magnetic field around an arbitrarily chosen pivot point, will not experience
any change of flux anywhere in the body. Thus no voltage and no eddy
currents are induced, since

Gφ
8= =0
GW
for all possible closed current paths in the conducting body.

25
*HQHUDORQHGG\FXUUHQWVDQGORVVHVLQ+RPRSRODU(OHFWURG\QDPLF
%HDULQJV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
This is illustrated in fig. 15 left where a cylindrical body is shown, but it
should be pointed out that the rotor could have any shape, and there would
still be no induced voltage in any part of the rotor.

So it is quite obvious that no eddy currents will be induced in concentric

operation. But aren´t they required? Without eddy currents there will not be
any force either.

The first and most important comment on this is that in concentric operation
obviously the rotor position is correct, and we do not need to move it away.
Thus no force is required to accelerate it to another position. However, we
need stiffness in order to keep the rotor concentric. But stiffness does not
require force, just a force derivative. So stiffness itself does not require eddy
currents to be induced when the bearing is centered. This will be further
analyzed later in this chapter.

Should there be static forces involved, like gravity, this force has to be
compensated for. It can be done by letting the bearing operate at a certain
excentricity, which is the normal thing to do for other bearings like
conventional fluid film bearings. This will induce eddy currents as will be
explained in the section on excentric operation. In other words,
electrodynamic repulsion is used to compensate for gravity.

The second comment on whether eddy currents are necessary or not during
concentric operation, is that there exists a better way of compensating for
gravity and other static forces than the method described above. By using
magnetostatic forces instead of electrodynamic forces, the rotor is allowed to
remain centered in the bearing. Magnetostatic forces from permanent
magnets have the advantage that they do not require eddy currents, so this is
the method to be preferred. Thus it can be concluded that eddy currents
related to static forces belong to the category of unnecessary eddy currents.

Until recently it was generally thought that magnetostatic forces would

destabilize the bearing and counteract the eddy current stiffness. This is not
necessarily true as will be explained in chapter 4.3, “Separation of lift force
and stiffness”.

There are two kinds of eddy currents that actually can be induced even when
the rotor is centered. One concerns unnecessary eddy currents due to space
harmonics caused by inhomogenous magnetic materials and badly machined
stator parts. The other kind of eddy currents, which is not necessary, but
highly advantageous, is damping currents that occur when the radial speed is
not zero. They will be offered a special chapter.

26
 ([FHQWULFRSHUDWLRQ&RPSDULVRQZLWKLQGXFWLRQJHQHUDWRUV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 ([FHQWULFRSHUDWLRQ&RPSDULVRQZLWKLQGXFWLRQ
JHQHUDWRUV
Necessary eddy currents are induced whenever dynamic forces cause the
rotor position to deviate from the center position. Their purpose is to push the
rotor back to its original position. To understand the physics behind these
currents we will compare the bearing with an induction generator.

Take a look at the bearing in fig. 15 right. When the rotor is not centered, the
flux, as seen from the rotor surface, is not homogenous any more, and eddy
currents are induced that repel the flux lines so that there seem to be a mirror
image of the stator magnets inside the conducting cylinder. However, the
imaging is very complex and phase delayed, so we definitely need a 3D-FEM
program to visualize and optimize the eddy current paths. The results from
these simulations are shown in several illustrations and diagrams in chapter 6.
A simplified eddy current path is illustrated and viewed from different angles
in fig. 15b and 16a.

ω × ×

×
×I


-
.
×
I+
.
.
I=0 ..

)LJDE3ULQFLSOHRIWKHKRPRSRODULQGXFWLRQEHDULQJ.




From fig. 15b it is obvious that the homopolar bearing rotor operating in an
excentric position will have many similarities with a two pole induction
generator operating at very high slip. This motivates the term induction
bearing.


27
*HQHUDORQHGG\FXUUHQWVDQGORVVHVLQ+RPRSRODU(OHFWURG\QDPLF
%HDULQJV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

ω B

)LJ D 3HUVSHFWLYH RI

WKH FRQGXFWLQJ F\OLQGHU LQ
ILJE


ωslip B

)LJ E  &RPSDULVRQ

ZLWK DQ LQGXFWLRQ
JHQHUDWRUURWRUFDJHZLWK
VOLS

In fig. 16a a perspective view of the conducting rotor cylinder is shown, and
in fig. 16b the corresponding rotor cage from an induction machine. In both
fig 16a and 16b the rotor has been displaced downwards as in fig. 15b, so that
the rotor will see a diametrical two pole flux component. In fig 16a and b the
homopolar flux component is removed for clarity, since it does not contribute
to any eddy currents.

In induction generators the forces from the currents have two force
components, one tangential drag force component )t producing torque, and
one normal component )n producing radial force. In the cage the conductors
are optimized to produce the largest possible breaking torque. In the
induction bearing we optimize for largest possible relation between normal
force and drag force. This is illustrated in fig. 17.

28
 ([FHQWULFRSHUDWLRQ&RPSDULVRQZLWKLQGXFWLRQJHQHUDWRUV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

× . × .
)t  ×
v )t  × v

)n
)
)n
)

)LJ&RPSDULVRQEHWZHHQ/RUHQ]VIRUFHVLQDQLQ
GXFWLRQJHQHUDWRUOHIWDQGDQLQGXFWLRQEHDULQJULJKW

A few words about normal forces has to be said in this context. Fig. 17 left
shows the normal and tangential components of the Lorentz force in an
induction generator with many poles. It could as well be a linear generator.
The Lorenz force is produced from the current in the rotor bar, the so-called
active part of the electric circuit. To the right the same is shown for an
induction bearing with many poles.

In the homopolar bearing the rotor will see only one pole, or two poles when
it is not centered, as described above. The curvature of these poles is so high
that the bearing is far from linear, and the direction of the forces becomes
less obvious. Actually, most of the Lorenz forces in an induction bearing
comes from the short circuit rings, which are represented by the ends of the
conducting cylinder. Compare the cylinder with the cage in fig. 16.

Furthermore, if iron is used in the rotor, as is always the case for induction
generators, there are also strong attractive forces present which we will
denote reluctance forces. They are strongly destabilizing in the radial
direction. On the other hand they also increase the inductance of the circuit,
which has a positive effect on the direction of the Lorentz force. The total
force is then the sum of the Lorentz force and the reluctance force.

For this reason iron is sparsely used in electrodynamic bearing rotors, though
it is often advantageous to increase the inductance. Filatov solved this
problem by setting the iron inductance apart from the magnetic circuit, which

29
*HQHUDORQHGG\FXUUHQWVDQGORVVHVLQ+RPRSRODU(OHFWURG\QDPLF
%HDULQJV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
is possible with a wire wound rotor. In the kind of induction bearing we are
studying in this report we have chosen a short circuit bulk conductor, a
cylinder, which can´t be series connected to a separate inductor, but has the
advantage of being able to operate at very high speeds.

30
 7KHSULQFLSOHEHKLQG]HURORVVEHDULQJRSHUDWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 7KHSULQFLSOHEHKLQG]HURORVVEHDULQJRSHUDWLRQ
We are now ready to formulate a very simple and yet maybe one of the most
beautiful principles in the realm of magnetic bearings. It explains in general
words how it is possible to suspend a heavy rotor in eddy current bearings
without the induction of any eddy currents at all.

†
)RUDQ\JLYHQRSHUDWLQJSRLQWLWLVSRVVLEOHDFFRUGLQJWRWKHVHSDUDWLRQ
FULWHULRQWRFRPSHQVDWHDVWDWLFORDGLQDQ\GLUHFWLRQZLWKDFRXQWHU
DFWLQJIRUFHIURPSHUPDQHQWPDJQHWVDORQHZLWKRXWWKHLQWURGXFWLRQ
RIDQ\LQVWDELOLW\

†
6WLOODFFRUGLQJWRWKH(DUQVKDZVWDELOLW\FULWHULRQDVWDELOL]LQJVHFRQG
GHULYDWLYHRIWKHPDJQHWLFHQHUJ\SRWHQWLDOLVUHTXLUHGLQDWOHDVWRQH
GLUHFWLRQDQGLWKDVWREHRIQRQPDJQHWRVWDWLFQDWXUH

†
$VHFRQGGHULYDWLYHRIDQHQHUJ\SRWHQWLDOLVDIRUFHGHULYDWLYH

†
(GG\FXUUHQWVFDQSURYLGHVWLIIQHVVZKLFKLVDIRUFHGHULYDWLYH
RIHOHFWURG\QDPLFQDWXUH

†
$IRUFHGHULYDWLYHGRHVQRWUHTXLUHDQRQ]HURIRUFH

†
:LWKRXWIRUFHWKHUHLVQRQHHGIRUHGG\FXUUHQWV

†
:LWKRXWHGG\FXUUHQWVWKHUHDUHQRORVVHV




When designing a magnetic bearing according to this principle, the bearing
will produce no eddy currents until some disturbance of the system takes
place that causes the rotor to move outside its operating position. All we need
to do is to choose an eddy current bearing that does not produce eddy
currents when they are not needed. Due to the rotational symmetry of the
homopolar bearing this is the ideal choice, as it cannot produce eddy currents
when center positioned.

31
*HQHUDORQHGG\FXUUHQWVDQGORVVHVLQ+RPRSRODU(OHFWURG\QDPLF
%HDULQJV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
4.3.1 Separation of lift force and stiffness
There are two very important aspects of the Earnshaw theorem. The first one
is referred to as the “Earnshaw stability criterion” and explains why three-
dimensional stability in a magnetically suspended system requires some
means of stabilization, as we have described earlier.

The second aspect is a special case of the normal formulation of the theorem,
and to the knowledge of the author it has is seldom or never before been
referred to. Thus we will call it the “Earnshaw separation criterion”. It makes
it possible to separate the stiffness from the lift force in a permanent magnet
bearing, PMB.

F=0 F>0

k>0 k=0
)LJ$[LDOEHDULQJWRWKHOHIWDQGD[LDOXQORDGHUWRWKHULJKW

Fig. 18 shows how this can be applied to axial bearings. To the left a
preloaded axial bearing, and to the right a bearing with large lift force, but
with no stiffness in any direction. We will denote the bearing to the right an
unloader. Unloaders can be used to improve the load capacity of any bearing
type without hazarding the stability. Thus the separation criterion can be
predicted to be of great importance in the design of new machines.

If equation (1) is reduced to

G 29 G 29 G 29
= 2 = 2 =0
G[ 2 G\ G]

it means that we have no stiffness, neither stabilizing nor destabilizing, in any

direction as long as we stay close to the working point. But we can still have
a large force, that is

32
 3HUSHWXXP0RELOH±KRZFORVHFDQZHFRPH"
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

G9 G9 G9
≠ 0 RU ≠ 0 RU ≠0
G[ G\ G[

where by definition

G9 G 29
= − )[ and = −N [
G[ G[ 2
where ) and N denotes the force and the stiffness respectively.

It is to be observed, that the stiffness is zero, while the force is non-zero. This
is the opposite compared to the stability criterion where the force is zero and
the stiffness in at least two directions can be chosen at will.

Thus, by combining one bearing of each kind, freely choosing the size of
each, it is possible to achieve any desired combination of stiffness and lift
force. In other bearing types like ball bearings this cannot be done, as in these
bearings the lift force is the product of the stiffness and the eccentricity.

Take a careful look at fig. 18. The two lower magnets are the same in the
bearing and in the unloader. The only difference lies in the upper magnet.
Thus instead of taking one bearing of each kind, as proposed above, only the
upper magnet needs to be changed in order to achieve any combination of lift
force and stiffness. 

 3HUSHWXXP0RELOH±KRZFORVHFDQZHFRPH"
In chapter 4 we have concluded that neither stiffness nor load requires losses.
It is not too far fetched to assume that we have invented the perpetuum
mobile. However, this is unfortunately not true. There will always be some
stray losses from magnet imperfections and air drag. Even in ultra high
vacuum there are some air drag losses that we can’t get rid of.

Furthermore the rotation of the earth will cause gyroscopic effects that has to
be dealt with, most methods causing small but nonzero losses.

Nevertheless, the invention of different kinds of homopolar bearings

definitely contributes to a large scientific step towards the ultimate perpetual
motion machine. It is very difficult, even to imagine, a physical phenomenon
that would create less losses at high speed than the homopolar bearing, even
worse to make an industrial product based on that concept.

Reports with superconducting bearings sometimes show impressive spin

down curves, but one has to remember that cooling of the bearings is rather
energy demanding.

33
'HVLJQRIWKHUDGLDOIOX[KRPRSRODULQGXFWLRQEHDULQJ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

5 Design of the radial flux

homopolar induction bearing
 *HQHUDOFRPPHQWVRQEHDULQJGHVLJQ
Since magnetic bearings differ much from conventional bearings like ball
bearings, it is important to define the purpose of the bearing before
dimensioning the bearing, otherwise we are likely to come up with a bearing
that is heavily over dimensioned.

The purpose of the induction bearing is to

• provide enough radial stiffness and damping to achieve rotor-
dynamic stability within a certain and predefined speed range,
• provide enough radial restoring forces when not centered to
compensate for dynamic load.

The bearing does generally NOT need to:

• provide any force when centered, since a permanent magnet
unloader better provides a static force,
• compensate for any unbalance force, since it is often better to let the
rotor rotate around its mass center than to compensate with large
radial forces to keep it in its geometric center,
• provide axial stiffness.

In appendix some simple hands on rules are given that are useful when
designing a new induction bearing. These rules are based on the analysis in
chapters 6 and 7.

It should be pointed out that a rotor dynamic simulation of the application in

question is as important as the design of the bearings, dampers and unloaders.
The design stage normally goes hand in hand with the dynamic simulations
during the development phase of a new product. Clearly, this dissertation is
not designated to rotor dynamics, but rather to provide engineers in rotor
dynamics with data in a form that can easily be adopted in such FEM
programs.

It is neither a topic of this dissertation to go further into the analysis of axial

bearings and axial and radial unloaders. What has been mentioned in chapter
3 forms the basis for the understanding of how separation of forces can be
done in order to utilize the induction bearing optimally. The following
description and analysis covers only the novel homopolar induction bearing.

34
 %HDULQJGHVFULSWLRQDQGGUDZLQJV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 %HDULQJGHVFULSWLRQDQGGUDZLQJV
The geometry of an induction bearing is comparatively simple. All parts are
rotationally symmetric, including the magnets. Fig. 19 shows a cut through a
“real” induction bearing for spindles. The picture is a photo realistic image
generated using 3D-CAD.

The four axially oriented magnets generate the magnetic flux. The number of
magnets can be optimized for the appropriate stiffness. Between the magnets
iron washers, which we will denote intermediate pole shoes, are placed to
concentrate the flux and change the direction so as to create a radial flux
penetrating the conducting cylinder.

The radial flux causes eddy currents to be induced in the conducting and
nonmagnetic rotor when it rotates in a non-centered position. Also the shaft,
which is inserted into the conducting cylinder, is preferably made from a
conducting and non-magnetic material, but it is not a demand.

The ferromagnetic end plates have a slightly different shape than the
intermediate pole shoes, but in the analysis a somewhat simplified shape will
be used, as in fig. 20. Their purpose is to avoid leakage flux and to increase
the inductance of the magnetic circuit.

The housing is made of non-magnetic material in order not to short circuit the
magnetic flux.

Conducting rotor
End plate

Housing

Magnets Iron pole shoes

)LJ'HVFULSWLRQRIWKHKRPRSRODULQGXFWLRQEHDULQJ

35
'HVLJQRIWKHUDGLDOIOX[KRPRSRODULQGXFWLRQEHDULQJ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 ,QQHUURWRUEHDULQJIRUVSLQGOHVDQGWXUELQHV
Fig. 20 shows a principle drawing of an inner rotor induction bearing with
two magnets. Compare with the bearing in fig. 19 that has four magnets. The
term inner rotor is used when the rotor spins inside the bearing stator. Other
rotor alternatives are found in the following subsections.

One difference from fig. 19 is the shape of the end plates. Fig. 20 shows how
they are normally modeled in the FEM software. The thickness of the end
plate is Ow while the thickness of the intermediate pole shoes, or washers, is
2Ow. Note also that the outer diameter of the washers normally is less than the
diameter of the magnets. This is to prevent too much leakage flux penetrating
the housing, where it does not contribute to bearing stiffness.

'RZ

GRZ

'U
WU GU

v]

OZ

[ B
O U 2OZ
\
OP

GRP

'RP

)LJ6SLQGOHEHDULQJQRPHQFODWXUH

36
 ,QQHUURWRUEHDULQJIRUVSLQGOHVDQGWXUELQHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

The thickness of the copper tube is chosen with regard to the skin depth.
Once the speed range has been determined, the thickness W U has to be at least
one skin depth calculated for the lowest operating speed, ω min .

'U − G U
WU ≡ ≥ δ VNLQ
max

2
where
2
δ VNLQ
max
=
ω min σµ 0

The length of the cylinder, Or, has a large influence on the end effect currents
and is analyzed in chapter 7. It turns out that there are two good solutions;
either the rotor should be slightly shorter, or much longer than the stator.

The magnetic length, or thickness, of the magnets, Wm, affects the speed range
of the magnets. Thick magnets are used for low speed operation, as will be
shown in the analysis.

Magnet width, Zm , defined byZm 'om±Gom, influences the radial flux
density in the airgap which in turn has a quadratic effect on stiffness. Since
magnets are expensive, an important task in chapter 7 will be to optimize
stiffness with regard to the amount of magnet material.

If the shaft is hollow, magnets can be placed inside the rotor as well. In the
following subsections we will describe two other types of induction bearings,
both using magnets on the inside of the shaft.

To distinguish magnets outside the rotor from magnets inside, the index “RP”
will be used for outer magnets, while the index “LP” will be used for inner
magnets.

Outer rotor bearings are used with outer rotor motors and are specially suited
for flywheel applications, fig. 3. That is why we prefer to denote them as
flywheel bearings. In an outer rotor design the rotor spins outside the stator,
which has several rotor dynamic advantages. Also the leakage flux from the
bearing magnets is reduced, making the bearing cheaper since less material is
required.

The most powerful bearing is a combination of the above. Placing magnets

both on the inside and on the outside of the rotor cylinder approximately
doubles the bearing stiffness. Such bearings are denoted intermediate
bearings.

37
'HVLJQRIWKHUDGLDOIOX[KRPRSRODULQGXFWLRQEHDULQJ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

5.3.1 Outer rotor bearing for flywheels

Applications like the flywheel in fig. 3 and the double flow compressor in
fig. 4 require an inside out motor/bearing concept, often referred to as an
outer rotor design. A bearing designed for this purpose is shown in fig. 21.

The main difference compared to the spindle bearing in fig. 20 is that the
diameters of the iron washers between the magnets are changed to reduce the
leakage flux on the inside of the stator magnets.

A mechanical advantage with this arrangement is that the carbon fiber

bandage required to prevent the conducting cylinder from exploding at very
high speed does not interfere with the magnetic flux in the air gap.

'U

GU

'LP,DLZ

GLZ

GLP

]

B
[
\

)LJ)O\ZKHHOEHDULQJQRPHQFODWXUH

38
 ,QQHUURWRUEHDULQJIRUVSLQGOHVDQGWXUELQHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

5.3.2 Intermediate rotor bearing for hollow shafts and

pumps
In some applications like turbomolecular pumps it is advantageous to use a
thin walled hollow shaft. Such a shaft enables fast acceleration, large
pumping surface and small gyroscopic effects.

Axial magnet
Conducting rotor Distance rings

Iron washers End plate Inner stator mounting rod

 )LJ'HVFULSWLRQRIWKHLQWHUPHGLDWHURWRUEHDULQJ

The bearing is shown in fig. 22. It should be noted that the nomenclature is
the same as for the previous two bearing types. This bearing comprises both
bearings, contains twice as many magnets and has approximately double the
stiffness.

An interesting aspect is the inductance, which is about twice as large as the

other bearing types. This means that losses are not twice as high as the other
bearings, as will be seen in the analyses presented in the following chapters.
Thus, this bearing is clearly the best choice, if the design permits this kind of
rotor.

39
%HDULQJDQDO\VLV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

6 Bearing analysis

 3XUSRVHRIWKHDQDO\VLV
A magnetic bearing is an advanced machine element and is to be applied by
machine engineers. However, the analysis of the bearing is more like the
analysis of electrical machines like motors and generators. Thus an important
goal of this analysis is to convert the electromagnetic properties of the
bearing to useful mechanical data.

 (OHFWULFDOSURSHUWLHV 0HFKDQLFDOSURSHUWLHV

Voltage Speed
Current Force
Current space gradient Stiffness
Current speed gradient Damping
Resistance Losses
Phase lag Force angle
Demagnetization Lifetime

The analysis will be performed in the following way. As a starting point the
magnetic circuit is investigated, without any rotating electric conductor. Then
the conductor is introduced into the airgap followed by an analysis of
possible eddy current paths that can be induced in the cylinder because of this
magnetic field during different types of bearing operations.

The paths are compared with currents induced in conventional generators,

and the circuits are modeled in such a way that conventional generator theory
can be applied. Resistance and inductance of these circuits differ somewhat
from the generator, and is thus analyzed in more detail.

When the magnetic flux and the currents are known the forces can be
calculated, and from these data properties like losses and stiffness can be
derived.

Finally the FEM analysis is performed in chapter 7 and the results are
compared to the analytical values derived in chapter 6.

40
 0DJQHWLFFLUFXLW
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 0DJQHWLFFLUFXLW
The magnetic circuit from the permanent magnet rings is illustrated in fig. 23.
Since the bearing is homopolar, it is convenient to introduce cylindrical
coordinates. In fig. 24 the radial flux component is plotted schematically
versus cylinder coordinate U at ] . Flux is defined positive coming out from
the iron pole shoe. The analysis will be performed for an inner rotor bearing.

The maximum radial air gap flux density %0occurs at the surface of the pole
shoe and decreases towards the center of the stator where it is zero.

]

B
[
U
\

)LJ0DJQHWLFFLUFXLW
%U
%0

Usable flux, ∆%U

Ja Wr JaJH

Gom/2 U
Operating interval, ∆Uop

)LJ5DGLDOIOX[GHQVLW\LQVLGHEHDULQJVWDWRU

41
%HDULQJDQDO\VLV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

Since the cylinder has a limited thickness we can only use a limited volume
of the magnetic flux. The operating interval ∆rop is defined by

∆URS = W U + 2 J D (6.1)

where J D is the available airgap defined by

JD = J − JH (6.2)

Here J H is the emergency airgap required by the emergency bearing.

Normally the emergency bearing clearance is set to half the airgap, thus
equation 6.1 becomes

∆URS = W U + J (6.3)

One of the most important quantities when calculating eddy currents is the
radial flux gradient δ%U / δU . Since ∆URS is small compared to Gom we will
assume that the flux gradient is linear in this region, thus

δ%U (U ) ∆% U (U )
= =
δU ∆U
G G (6.4)
% U ( RP − J H ) − %U ( RP − J H − ∆URS )
= 2 2
W U + 2J D

For the outer rotor bearing the analysis is similar, but for the intermediate
bearing, fig. 22, the analysis is simpler, since the inner magnet helps linearize
the flux in the whole airgap. The gradient is also higher for the latter bearing
than for the other bearings, since the null radial flux limit lies approximately
in the middle between the magnets. Thus

δ%U (U ) − 2 %0
≈ (6.5)
δU WU + 2 J

It should be noted that for the intermediate bearing, %0 is lower than for the
other two bearing types, because the repulsive poles weaken the flux to some
extent.

42
 0DJQHWLFFLUFXLW
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

(GG\FXUUHQWV

6.2.1 Eddy currents induced due to eccentricity

When the rotor is centered, as in fig. 26a, there are no points on the
conducting cylinder surface that can see any change in flux density as it
rotates in the homogenous magnetic field. Thus according to Lenz’s law there
can be no induced currents.

In fig. 26b the rotor is displaced downwards in the negative y-direction and
the corresponding eddy currents are shown in fig. 26d. A perspective view is
given in fig. 25. In order to improve the viewing angle the coordinate system
has been turned so that the y-axis points to the left. Accordingly, the rotor is
displaced to the right. A perspective view of the bearing and the currents are
shown in fig. 28.

Since the shape of the eddy currents by no means is obvious, we shall

analyze them a bit further. Before doing so we need to introduce a coordinate
frame x1y1z1 translated in the xy plane by the vectors ∆U0 , fig. 25. We also
need a rotating coordinate frame xryrz1 fixed to the rotor in which we will
calculate the eddy currents. The angle between [1 and [r is ωW. For
convenience we introduce a second cylindrical coordinate frame ρ φ z1 where
φ is defined as the positive angle between [1 and ρ .

ω
] ]1
[ [1, xr A B
•
ρ
\\1\r φ

∆U0

)LJ'HILQLWLRQRIFRRUGLQDWHV\VWHPVDWWLPHωW 

43
%HDULQJDQDO\VLV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

Now let us focus on point A in fig. 26c, which is spinning around axis ]1 at
speed ω. When the rotor is not centered, point A on the rotor surface will
experience not only the homopolar flux, but also a superposed diametrical
two-pole flux. Point A will thus see an almost sinusoidal flux with a constant
bias.

To find the eddy current amplitude and phase we need to know the time
derivatives of the normal flux penetrating the surface. Let us first make a few
simplifications concerning the eccentric motion of point A.

Assume a position for point A so that φ   at time W   Also assume a

certain displacement ∆Uo in the negative y-direction. At this instance of time
the position of A expressed in the stator coordinate system xyz is (ρ, -∆Uo,).

ω ω
G%U
=0 ε>0
GW
, =0

\

)LJD )LJE
]• [

∂%U ∂%U
>0 <0
∂W ∂W
ω x
\1 I-
• F
φ ]1 [1 I+
ρ A .
•

)LJF )LJG

44
 0DJQHWLFFLUFXLW
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

Since ∆U0 << ρ we can assume that U ≈ ρ for ωW andωW π.

With the simplification above we can write the radial position for point A as
it spins around axis ]1 as

U ≈ ρ − ∆U0 sin(ωW ) (6.6)

or

U ≈ ρ − ∆U0 sin(ωW ) (6.7)

Now the time derivative of the flux from an intermediate bearing, using
equation 6.5 and 6.7 is

∂%U ∂%U ∂U − 2 %0 2ω%0 ∆U0

= ⋅ ≈− ⋅ ∆U0ω cos(ωW ) = cos(ωW ) (6.8)
∂W ∂U GW W U + 2J WU + 2J

A typical flux distribution is shown in fig. 27. It has been calculated using the
Mega software, a program that we will use frequently in chapter 7.

From the flux time derivative it is possible to analytically calculate the

current distribution using the electrodynamic diffusion equation. However,
since the focus of this report lies in FEM calculations to govern visualization
and basic understanding of the fields and currents, we will not use this
approach here. Instead we will compare with well-known generator theory.

)LJ$LUJDSIOX[GHQVLW\SHQHWUDWLQJWKHURWRUVXUIDFH

45
%HDULQJDQDO\VLV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

In chapter 4 we compared the induction bearing with the induction generator due
to their obvious geometrical similarities. Thus we expect the conducting bearing
cylinder to behave like the induction generator cage in fig. 16b. Actually, we can
make an even better comparison with the synchronous generator, since the stator
frequency is zero and slip frequency equals the rotor frequency.

Thus we can calculate the axial eddy currents Lz in the rotating coordinate system
as if they were generator rotor bar currents being part of a sinusoidally
distributed 3-phase winding.

( ∂Φ U ∂W $ ∂%U
L] = = = (6.9)
5 2 + ( Mω/) 2 5 2 + ( Mω/) 2 5 2 + ( Mω/) 2 ∂W

where A is the pole area enclosed by each eddy current as defined in fig. 30.
Using equation 6.8 and 6.9, expressing ∆U0 in terms of eccentricity HJ,
eccentricity times airgap, the final expression for the current amplitude becomes

2ω$%0 HJ
Lˆ] = (6.10)
(W U + 2 J ) 5 2 + (ω/) 2

The axial current Lz turns direction into a tangential current Ltat the cylinder ends,
or, if many magnets are used, at each symmetry line B-B. The tangential current
in the bearing cylinder corresponds to the current in the short circuit ring in an
induction generator.

The phase shift, or load angle as is the common notation for generators, is then

5
θ = arctan (6.11)
ω/
Equation 6.10 represents a current that initially increases linearly with frequency
ω and then asymptotically reaches its maximum value. This is in agreement with
the results from the FEM-simulations shown in fig. 49.

The phase shift in equation 6.11 starts from π/2 at low frequency, which then
asymptotically reaches zero at high speed. This explains the angle in fig. 26d.

An attempt to illustrate the eddy currents is made in fig. 28-29. Fig. 28a and 29
shows the two main eddy current circuits at low speed, and fig. 28b shows the
theoretical current distribution at infinitely high speed.

46
 0DJQHWLFFLUFXLW
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

∆U
ω

ω = inf

)LJDE5HVWRULQJHGG\FXUUHQWFLUFXLWV

47
%HDULQJDQDO\VLV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

The pole area A, which is used in equations 6.9 to 6.11, refers to the area
where the flux is radial and which is enclosed by the eddy currents. The
currents will basically follow the geometry of the magnets, which makes it
easy to estimate an approximate value. In reality the current distribution is
not discrete as illustrated in fig. 28 and 29, but sinusoidally distributed. Thus
obtaining exact values for the area is not so simple. An expression based on
the current path in fig. 30 that is valid for inner, outer as well as intermediate
bearings, all with iron poles, is

πG UF
$ ≈ 2OZ ( − Z3 ) (6.12)
2
where Grcis the diameter of the rotor cylinder centerline and Z3 is defined in
equation 6.18.

For a bearing without pole shoes the area is not that well defined, and the
direction of the flux lines have to be carefully studied. A rough estimation
would be to exchange Ow in equation 6.12 for Om in this case.

)LJ(GG\FXUUHQWVLQDIO\ZKHHOEHDULQJ

48
 0DJQHWLFFLUFXLW
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

Finally we need to find an expression for the A B

resistance R and the inductance L for the
eddy current paths. The simplified model in
fig. 30 can be derived by making some
simple assumptions, but one has to bear in
mind that the currents are not bound to
discrete paths. For future work, in order to
develop accurate analytical formulas, we
need to solve the diffusion equation. But at
this stage of the development it feels more
appropriate to use the generator model, as it
will give a better understanding of the
physics than the typical series expansions
resulting from a solution to the diffusion
equation.

The shaded eddy current path in fig. 30 has A B

been linearized in fig. 31. This simplification
will actually give a result that is somewhat
too high due to neglected corner effects. )LJ

The total resistance 5 equals

ρ Z1 Z1 + Z3 Z3
5 = 2( 51 + 2 52 + 53 ) = 2 ( + + ) (6.13)
WHII O1 O2 O3

where

 W , WU < δ VNLQ
WHII =  U
δ VNLQ , WU > δ VNLQ

O1 O2 O3

Z1 R1 R2 R3 Z3
Weff

)LJ/LQHDUPRGHORIRQHHGG\FXUUHQWSDWK

49
%HDULQJDQDO\VLV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

The lengths and widths can be found geometrically, in a somewhat simplified

form, from fig. 30.

 Z3 O
O1 = πU − 2( 2 + U2 ) where U2 = P
4
 2πU2 πO P
O 2 = =
 4 8
O 3 = 2O Z (6.14)
 OP
Z1 =
 2
Z3 = U 2
 π
In these formulas U refers to the radius of the center of the rotor cylinder, thus

5U + UU
U = UUF = (6.15)
2
The width Z3is by no means obvious and requires an explanation. We could
have chosen Z3 to be one quarter of the circumference of the rotor, which is
true in that aspect that the whole area actually carries eddy currents. There
exists no “white area” as in fig. 30 where the current is absent. We already
know that the current density is sinusoidally distributed, which means that the
main part of the currents in region 53 passes through the upper and the lower
part of fig. 30. What we want to do is to calculate a certain model width Z3
that gives us the same losses in this region, provided a constant current
density, as we would have got choosing Z3 to cover the whole area, fed with
a sinusoidal current density. Thus the total currents have to be equal, that is:

, ] = ∫ M] ⋅ G$
$
π /2
, ] = WHII ∫ ˆM sin α U Gα
0
]

, ] = ˆM] UWHII (6.16)

where α is an integration angle which is zero on top of fig. 30 and follows the
circumference of the rotor. This angle is π/2 turned counterclockwise from
the load angle θ.

50
 0DJQHWLFFLUFXLW
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

Also the losses shall be equal, thus

53 , ]2 = ∫ ρ M] G9
2

9
π /2
O
ρ 3 , ]2 = ρ ∫ ( ˆM sin α ) 2O3WHII U Gα (6.17)
WHII Z3 0

Solving, using equation 6.16, gives

2U
Z3 = (6.18)
π
Summing up, we find the resistance to be

2U O P
πU − −
5=
ρ
(4 π 2 + πO P + 2 πO Z ) (6.19)
W HII OP 2U O P U
+
π 2
Calculating the inductance / is strongly dependent on whether we use iron
pole shoes or not. If they are not used the inductance can be calculated as an
air wound coil. Filatov9 whose coils are similar and also air wound, uses the
formulas for the inductance from an infinitely long conductor, applied to the
tangential parts of the current circuits, as

µ0 1 O
/=O ( + ln 0 ) (6.20)
W
π 2 U0

where Otis the length of the tangential sections of the coil, which in fig. 30 are
the vertical parts of the current paths and which we approximate with O1U2.
The radius U0 of the conductor is approximated to Om/ and the distance
between the centers of the conductors O0 is Om/Ow where Ow in this case is the
eventual distance between the magnets as iron washers are not used.

51
%HDULQJDQDO\VLV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

6.2.2 Eddy currents induced by lateral motion

Rotor vibrations with frequency ωv will induce damping eddy currents in the
rotor similar to the restoring currents above. The vibrations cause the rotor to
se a change in flux proportional to its radial velocity, fig. 32, thus damping
will occur even at no rotational speed. At non zero speed ω the induced
currents will be dragged along with the rotor surface due to the phase delay.
Thus the damping force will have a side component like the restoring force.
We can call it a “phase delayed internal eddy current damping”.

It is to be noted that the damping circuit has a π/2 forward shift compared to
the restoring circuit. Thus the damping side force will be oppositely directed
and thus has a tendency to introduce a negative whirl, which will to some
degree counteract the forward whirl that the restoring force will try to
introduce.

The characteristics of the damping effect are like that of a viscous damper,
the side force not taken into account. In such dampers the force is
proportional to the vibrational speed.

Causes for vibration may be external loads, unbalance, whirl, gyroscopic

effects etc. Unbalance is of particular interest since vibrational deflection due
to unbalance is synchronous with speed, so the ratio ωv/ω = 1. No point on
the rotor will thus se any change in flux. Thus there are no eddy currents and
no additional losses. Since there is no force, absolutely no unbalance
vibration can be transmitted to the housing!

When using active magnetic bearings on unbalanced rotors it is important to

filter away the synchronous frequency from the controller. With induction
bearings this filtering is done automatically.

Compared to the circumferential speed of the rotor, the lateral speed is

normally very low, and the damping effect can thus almost be neglected.

52
 0DJQHWLFFLUFXLW
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

ω=0 vv

ω=w v

)LJDE,QWHUQDOGDPSLQJFLUFXLW

53
%HDULQJDQDO\VLV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

6.2.3 Stray eddy currents

Any inhomogenity in the permanent magnets, it may come from sintering or
from magnetization, will most likely cause a non rotationally symmetric
magnetic field, which in turn will induce stray eddy currents. These currents
will have a shape depending on the inhomogenity. The most common one is
misalignment between the geometric and the magnetic axis. Such simulations
will be performed in chapter 7. Local flux dips, or random change of
magnetic orientation around the perimeter of the magnet rings are also
common.

Stray currents may also result from inhomogenities or remanent magnetic

flux in the iron pole shoes. Iron screws and pins close to the airgap can also
alter the flux so as to induce unwanted eddy currents.

Lots of FEM simulations on different kinds of inhomogenities have been

performed, some of them shown in chapter 7, and they all show that the
influence on restoring forces is very limited, whereas the influence on losses
is large.

When ordering magnets it is very important to specify the maximum allowed

misalignments and local flux dips depending on application loss
requirements. Today not many magnet manufacturers undertake such
considerations.

Though, even in the case of not perfectly manufactured magnets, the

induction bearing still produce very low losses compared to many other
magnetic bearing types.

54
 %HDULQJIRUFHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 %HDULQJIRUFHV
This far we have analyzed the magnetic flux and the eddy current circuits. By
forming the product -î% along the eddy current circuit in fig. 30, the Lorenz
force distribution is achieved. In fig. 30 it is important to note that the
magnetic flux is radial in cut A-A while it is axial in cut B-B. This comes
automatically from the flux lines in fig. 23. In fig. 33a and b this is illustrated
from another viewing angle. Thus, in the linear model of the current path in
fig. 31, the current is tangential and the flux is axial in region 1. In region 3
the flux is radial and the current is axial. Since the length l1 normally is much
longer than the length l3, the main force is generated in region 1. In fig. 33c
below the force distribution from region 1 is illustrated.

.
B+ Bz-
z ×
× - × ×
I
. × It × .
I+ × ×
.
. × .
.

)LJD6DPHDVILJD )LJE&XW%%D[LDOIOX[
&XW$$UDGLDOIOX[
\

. θ
. × ) )res
× × ω
. × ) × . ∆U0 +
[
× 0z
. × .
.

)LJF-î%IRUFH )LJ G 5HVXOWDQW IRUFH DQG

GLVWULEXWLRQ EUDNHWRUTXH

55
%HDULQJDQDO\VLV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

Region 1 would in an induction generator refer to the short circuit ring, and
region 3 would refer to the active part or the rotor bars. For an electrical
engineer it might be a little bit confusing that the short circuit ring plays such
an important role in the induction bearing.

The resultant force )res in fig. 33d is obtained by integrating the force
contributions over the whole cylinder. The force angle θ is the same as the
electrical load angle θ, the phase shift defined in chapter 6.2.1. The phase
shift depends on speed, resistance and inductance.

If we know the point of attack, or any other point on the line defined by the
vector )UHV , the brake torque 0z can be calculated. Due to symmetry this line
has to pass the center of the homopolar flux, which is the origin of the stator.
Filatov1 has in his Lemma no. 1 shown that this is always true for an axial
homopolar bearing, and in a similar way it can be shown for radial flux
bearings as well.

If the rotor is displaced in the negative \- the torque can easily be calculated
as the product of the x-component of the force, )x with the displacement ∆U0.

If the bearing is not strictly homopolar, like for instance if the magnets are
not perfectly axially magnetized, the point of attack is somewhere else and
the brake torque will be larger.

In fig. 28a, b the currents are plotted for different speeds and thus different
phase shift. In fig. a the speed is low and the load angle is large. See also fig.
34. Thus the force angle is also large and the bearing force has a large side
force component, which has a strongly destabilizing effect on the rotor. In
fig. b the speed is infinitely high, and the force angle is 0 so that no side force
component is present. Thus the force is totally restoring.

Unfortunately we can´t even come close to infinite speed, so there will

always be a certain side force component. However, by using iron pole shoes,
or whiskers, between the magnets, the inductance is increased and thus the
relationship between restoring force and side force is increased. Anyway, it is
always necessary to carefully study the stability criterions for the complete
rotor setup.

Let us now calculate the force from one magnet using the simplification that
the force contribution comes solely from the tangential currents and that the
flux is axial in this region.

56
 %HDULQJIRUFHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

F B
θ
\\1

)LJ  5HVXOWLQJ IRUFH EHWZHHQ HGG\

FXUUHQWVDQGPDJQHWV

The force acting on the rotor can be calculated as the volume integral of the
Lorenz forces. One eddy current circuit covers two quarters of a magnet, thus
the integral has to be multiplied with a factor of 2, and the result is valid for
one intermediate magnet, when using the current I and the resistance R from
equations 6.10 and 6.19. For the end magnets the values are lower due to the
reduced flux density and the fact that the resistance is dependent on the
length of the rotor. The force acting on the rotor from one intermediate
magnet is the sum of 4 tangential current sections, thus

) = 4∫ - × % G9 (6.21)
9

The integral can be simplified using the fact that we know the load angle.
The force has to be parallel, so we can introduce an angle α measured from
the load angle and integrate from -π/2 to π/2. Projecting the force to this line
by multiplying with cos α gives us the integral

π /2
ˆM W cos α % ] cos α O P W HII U Gα
 ) =2
π
∫
− /2
2
(6.22)

where jt is the tangential current density which has its maximum at α = 0. At

this point the total current I equals

57
%HDULQJDQDO\VLV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

O
 , W = ˆM W W HII P  (6.23)
2
which is the same current as in equation 6.16. Thus equation 6.22 can be
written as
 ) = 2 π , W %] U  (6.24)

The force from the end magnet is much depending on the shape of the end
plate and the length of the rotor. But as we can see from the FEM-simulations
the two end magnets contribute approximately as much as one intermediate
magnet. Thus we can write the total force acting on a bearing with n magnets
(rows) as:

) ≈ (Q − 1) 2π , W %] U (6.25)

The force direction is along the load angle, which is given by equation 6.11.

58
 /RVVHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 /RVVHV
The losses for one intermediate magnet are the sum of two eddy currents.

3 = 2 5, 2 (6.26)

and the approximate formula for a bearing with n magnets is then

3 ≈ 2 (Q − 1) 5, 2 (6.27)

 6WLIIQHVV
Stiffness is defined as the space derivative of the force. If the rotor is
displaced for instance along the y-direction, the stiffness is defined as

G)
. =− (6.28)
G\

Filatov defines this stiffness as the “in plane” stiffness, since the force is
acting in the plane, not necessarily in the y-direction.

From the force equation 6.24 and the current equation 6.10 it is clear that
stiffness is constant with regard to displacement. The final expression for
stiffness is then derived by multiplying equation 6.10 and 6.25 and dividing
by the displacement, preferably expressing G\ in equation 6.28 in terms of
eccentricity, so that G\ = HJ.

 /RDG5DQJH

When the eccentricity is too large emergency bearing contact will take place,
depending on the actual radial position of that bearing. This limits the
maximum permitted dynamic load. As we have noted in chapter 5 there is no
limit for the static load carried by the separate permanent magnets, but the
maximum dynamic load carried by the eddy currents is then set to the
stiffness times the maximum allowed displacement.

59
%HDULQJDQDO\VLV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 5RWRUG\QDPLFV
Let us study a few aspects on rotordynamics, to make it easier to compare
induction bearings with other bearing types.

A bearing is said to be isotropic if the stiffness is the same in all directions.

Tilt pad bearings are for instance not always isotropic. Neither are all active
magnetic bearings. Homopolar bearings are.

High stiffness is required in a tool spindle in order to obtain accuracy, while

low stiffness is necessary in a centrifuge to avoid vibrations due to imbalance
forces. Basically a shaft rotating at a certain speed rotates about its geometric
center if the stiffness is high, while at low stiffness it rotates about its center
of gravity. A rotor accelerating from stand still to high speed will pass
through the first rigid body critical speed, which is the speed where the rotor
changes from rotating about the geometrical axis to rotating about its center
of gravity. Below the critical speed the rotor is said to run subcritically, and
above that speed it runs supercritically.

The first critical speed ω for a rigid rotor with two flexible radial bearings
equally distributed from the center of gravity, the bearings not sitting to close
to each other, can be calculated from

2.
ω= (6.29)
P

where m is the mass of the rotor. At this frequency the rotor vibrates so that
the axis of rotation is always parallel to the original direction. This mode is
called the cylindrical mode.

If the speed is increased another mode will be introduced, the so-called

conical mode. The names of the nodes refer to the shape of the volumes that
are built up by the vibrating shaft. The vibrational frequency for the conical
mode depends not on the mass of the rotor as for the parallel mode, but on
the moment of inertia in the x and y-directions. If this moment is very high
the conical mode will appear before the cylindrical mode.

Normally a rotor is not as symmetric as described above. This will most

likely result in two conical modes, one for each bearing.

At high speed things get more complicated. The two conical modes are split
up into four modes due to gyroscopic effects. Depending on factors like rotor
diameter and stiffness several elastic body modes will also be added. A
convenient way to represent the different possible frequencies is to map them

60
 5RWRUG\QDPLFV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

in a diagram like in fig. 35. In this diagram the operating line can be drawn,
and each crossing with a line referring to a specific vibration mode will
represent a critical speed. Some of these bending modes are more likely to be
introduced than others.

If a mode is introduced, the vibration level is normally very high, unless the
damping is very good or the acceleration or retardation is high.

Active magnetic bearings offer a very interesting way to eliminate this

problem, as the stiffness can be instantaneously altered, as i fig. 36. It is thus
possible to temporarily lower the stiffness before passing through a critical
speed so that the critical speed is reduced under the actual speed. When the
speed is higher the stiffness is raised again. In addition, with active magnetic
bearings the damping can be individually programmed for each bending
mode for maximum system performance.

Induction bearings offer another possibility to avoid the problem. As the

stiffness gradually increases from zero to an asymptotic value when the speed
is increased, it is possible to design a bearing so that the rotor always runs
supercritically, even at considerably low speed. We call this “resonant free
operation”. Fig. 37. shows how this is possible.

f
[s-1] 2:nd conical

1:st conical

Operating line

Speed, ω [s-1]


 )LJ0RGDODQDO\VLVRIDULJLGXQV\PPHWULFDOEHDP

61
%HDULQJDQDO\VLV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

f 2:nd conical
[s-1]

1:st conical

Speed, ω [s-1]

)LJ$FWLYHFRQWURORIILUVWFRQLFDOPRGHV



f Operating range
[s-1]

2:nd conical

1:st conical

Speed, ω [s-1]

)LJ5HVRQDQWIUHHRSHUDWLRQZLWKLQGXFWLRQEHDULQJV

62
 $X[LOLDU\EHDULQJV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 $X[LOLDU\EHDULQJV
During start up and landing induction bearings does not provide enough
restoring forces to keep the rotor centered. Thus some form of auxiliary
bearings are needed, sometimes called “landing bearings” or “touch down
bearings”.

Normally the induction bearing have integrated unloading bearings, so the

resulting force acting on the landing bearings is virtually zero. However, in
all kinds of magnetic bearings there is always a risk for whirl whenever a
touch down bearing contact occurs. A fully developed whirl will cause very
large radial forces. Thus careful considerations must be taken in order to
adjust damping and geometry of the landing bearings so as to prevent this
motion.

The author has, during earlier tests of heteropolar induction bearings,

successfully used rubber mounted PTFE bushings as landing bearings.
Correctly tuned no whirl is introduced, and thus no radial forces are apparent.
As wear is proportional to the product of surface speed and the normal
component of force, this means that wear is very low.

An interesting alternative to landing bearings is to have bearings in

continuous operation, like airfoil bearings. They can provide damping and
shock absorption and contribute to rotordynamics in a better way than
seldomly used landing bearings. At the same time the induction bearings
reduces the take off speed of the foil bearing, the latter normally referred to
as a great drawback when foil bearings are used alone. This bearing
combination is ideal for applications where vibrations and heavy shocks are
likely to occur like in vacuum cleaners.

The induction bearing can also be used in a vacuum environment, thus

allowing for very high-speed operation like in flywheels and gyroscopes. In
this case air bearings are not used. Instead coated plain bearings equipped
with a centrifugal clutch is more appropriate.

 &RPSDULVRQZLWKEDOOEHDULQJVWDQGDUGV
As many parameters are interdependent, data are usually given for certain
reference loads. For a ball bearing for instance one reference load is the
dynamic load-carrying capacity, which according to ISO 281:1990 is defined
so that the probability is 90 % that the bearing lifetime will exceed 1 million
revolutions.

Another reference load is the static load-carrying capacity, which according

to ISO 76:1987 is defined based on the maximum contact pressure.

63
%HDULQJDQDO\VLV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

According to the SKF bearing catalog, this load results in a permanent elastic
deformation of the contact surfaces that is approximately 0.0001 of the
diameter of the ball.

The static load-carrying capacity is only valid for very low speed applications
and thus has no relevance in a high speed machine.

As a result from the elastic deformation, the inner ring of a ball bearing will
rotate in an eccentric position. Also, for other conventional bearing types,
there is a certain eccentricity directly proportional to the bearing load and
inversely proportional to the stiffness. Thus when heavy load is applied, high
stiffness is required in order not to get too large deflections.

This can sometimes be contradictory to results from rotordynamic

calculations where lower stiffness may be required for smooth operation.

As was described in chapter 5 a magnetic bearing offer the unusual

possibility to separate the load from the stiffness so that the eccentricity can
be chosen at will. Normally the eccentricity is then set to zero during normal
operation.

64
 6LPXODWLRQPHWKRG
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

7 FEM -simulations
In chapter 5 the different bearing topologies were discribed, and from chapter
6 we have gained a basic understanding on how the eddy currents are induced
and what bearing forces they give rise to. In this chapter, which is the main
and most important part of the disertation, the currents and forces will be
analysed using a 3-dimensional finite element method, 3D-FEM.

 6LPXODWLRQPHWKRG
The software we will use for the simulations is called MEGA, a 3D-FEM
program developed by the University of Bath,
.
A typical printout from the program is shown in fig. 38, where the current
density is represented by coloured areas and the bearing force distribution is
represented by arrows.

)LJ0HJDZLQGRZ

65
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

Since 3D-FEM calculations are very time consuming, and this dissertation
presents a survey of a vast number of calculations performed during a six
year period, much effort was put into selecting the software, and building fast
Linux servers.

7.1.1 Software
As this type of bearing configuration represents a true 3-dimensional problem
it could not be solved with any of the 2D programs available on the market in
1995 when the project started. Earlier studies by the author on null flux
bearings as described in chapter 2, were performed using the 2D software
ACE from ABB, but it could not be used for this task.

The choice fell on Mega, at the time a recently developed 3D FEM program.
It was the only identified FEM program that had implemented a Minkowsky
transformation.

In 1995, time step solutions was the trend in 2D simulations, but in 3D

simulations they turned out to be all too time consuming. The Minkowsky
transform makes it possible to solve the rotating field problem in one single
calculation.

The Minkowsky transformation increases the number of equations in the

eddy current elements by a factor of two. Since about 30% of the elements
are used to build up the mesh of the conducting rotor, this means tha the total
number of equations are increased approximately 30%, which in turn
increases the calculation time 69% compared to the time to solve a single
time step. Since a large number of time steps are required, the time saving
using the Minkowsky transformation instead of time steps is approximately
one day per simulation.

In normal configuration Mega is not optimized for 30% eddy current

elements, so the team at Bath had to modify the software used for these
calculations.

7.1.2 Platform
Mega is now run on a 32 bit Linux server using 1 GByte RAM memory, dual
550 MHz pentium processors and 2 parallell fast SCSI hard disc drives. This
is the fourth Linux sever built during the project. The maximum allowed
number of equations in this configuration is 250 000. To solve more
equations a 64 bit processor would be needed. Normally the model size is
150 000 equations which takes about one hour to solve as long as hard disc
swapping is not required.

66
 0RGHOJHRPHWU\RIWKHVWXGLHGEHDULQJV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 0RGHOJHRPHWU\RIWKHVWXGLHGEHDULQJV
All studied bearings have a rotating copper or aluminium cylinder. In the
stator there are one or more permanent magnet rings. The stator magnets are
located on the outside and/or on the inside of the cylinder as described in
chapter 5. Thus we refer to these bearings as inner rotor or outer rotor
bearings. In case there are magnets both inside and outside the conducting
cylinder we refer to such bearings as intermediate bearings. Fig. 39 shows a

Outer magnet

Conducting rotor )LJ

Inner magnet

simple intermediate bearing without pole shoes as drawn by the Mega painter
subroutine. Due to axial symmetry around plane B-B only one half of the
bearing is required to be modeled, fig. 40.

)LJ B

67
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 6\PPHWU\SODQHVDQGERXQGDULHV
In order to reduce the mesh size the following symmetry conditions have
been used:

Symmetry plane A-A in fig. 41 devides the bearing axially in two identical
halves. The appropriate boundary conditions for this plane are “tangential
flux” and “normal currents”. The use of this symmetry should not give rise to
any calculation errors.

Symmetry planes B-B and C-C in fig. 41 are used when calculating bearings
with more than three poles. They are no “true” symmetry planes and will give
rise to minor calculation error. The larger the number of poles the smaller the
deviation. For a theoretical bearing with an infinite number of poles the
values should be correct as the flux at the planes are not affected by end
effect leakage. The boundary conditions in plane C-C are chosen the same as
in A-A. In plane B-B the flux lines are perpendicular.

In order to reduce calculation errors for bearings with a limited number of

magnets the end effects are calculated separately. Thus for example a bearing
with 4 identical magnets as in fig. 41 is calculated as the sum of two ends and
four magnet halves. The ends, which are represented by the axial distance
between plane C-C and D-D, are calculated in one model, and the magnet
halves represented by the distance between A-A and B-B are calculated in a
separate model. The first model also includes some levels of air in order to
properly deal with the end effects.

Both models are surrounded with air in the radial direction in order to avoid
unnatural flux concentration. The side boundaries are “zero A”, that is the
field is tangential as above so that no field can escape from the model.





)LJ0HJDPRGHOV\PPHWU\SODQHV

68
 0HVKJHQHUDWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 0HVKJHQHUDWLRQ
A very important aspect of this work is to find a way to create a mesh, which
will generate results that are accurate enough. The reason why this is more
important for an induction bearing than for a motor or other type of
electromagnetic machine is that the losses in the induction bearing are almost
zero. Furthermore, post processing of some of the data involves methods
leading to cancellation of several significant numbers.

7.4.1 Mesh generation in Mega

Mega differs from many other FEM programs in which the way the mesh is
generated. Basically nodes and elements are created to start with, elements
being either triangular or quadratic. Initially a two-dimensional mesh is
generated, which later is extruded into a three-dimensional mesh. After
extrusion the mesh can be modified or skewed to any desired shape. When
the mesh is ready, material properties are associated with each element. In
Mega materials are denoted regions. A list of the regions used is given in
chapter 7.5.

7.4.2 Comments on automatic mesh generation.

It is possible to have the mesh elements generated automatically in Mega.
However, as we are designing a bearing with almost zero losses it is utterly
important to make a symmetric mesh. When doing parametric changes to the
geometry, we want to be sure that the change in the calculated forces is a
result from the change in the parameters and not from the change in a newly
generated mesh.

7.4.3 Square or triangular elements?

In Mega it is possible to use and combine square and triangular elements. In
triangular elements the variables are kept constant, whereas in the square
elements they are linearized. We found that we could not use any triangular
elements in or close to the eddy current regions, otherwise additional eddy
currents would be induced purely related to the bad mesh and with no
physical significance.

7.4.4 Structure of the manually generated mesh

In addition to what has been said above, about the importance of the mesh
being symmetric, there are some other considerations necessary to be aware
of as well. To obtain highest possible accuracy, for a given and limited
number of equations possible to solve, it is necessary to find a good
compromise between the number of sectors, layers and levels.

69
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

We found an inversely quadratic dependence on stray losses as a function of

sector width, while the number of levels had only limited effect on the
accuracy. The number of layers on the other hand has to be at least four in the
air gap, and in the conducting cylinder there has to be a sufficient number of
layers to deal with the penetration depth.

Furthermore the mesh elements, especially in the air gap, shall not be to flat.
Mega can handle aspect ratios as high as 80, but it is strongly recommended
to use ratios of less than 3. We have tried to keep all elements as close to a
quadratic shape as possible, fig. 42 to 44.

Close to the airgap it is further recommended that there is a gradually

increasing size in mesh elements.

As can be seen from fig. 44, we have not allowed any triangular elements
close to the airgap, neither in the magnets nor in the conducting cylinder.

Taking the above discussion into consideration, we estimated that a good

compromise was to focus on keeping the sectors narrow, 1.25 degrees each in
the airgap, and reducing the number of levels. 15 levels or less were used for
most models. We used 4 layers of square elements in each airgap, and
between 6 and 10 layers in the conducting cylinder depending on rotor
thickness.

Lots of air is required

around the magnets.

Tangential flux boundary

conditions

)LJ0DQXDOO\JHQHUDWHGV\PPHWULFPHVK

70
 0HVKJHQHUDWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

The rotor has been translated in the airgap in the negative y-direction when
forces due to eccentricity are calculated. Sometimes, as in fig. 43, it has been
more convenient to move the stator in the positive y-direction. The result
should be the same if the boundaries are far from the rotor, which requires
that there is a lot of air outside the magnets. Air is also needed in the z-
direction outside the end magnets, which means that we need some levels of
air there as well.

Rotor displaced
downwards

)LJ
\

[

Four layers of air, as in fig. 44, are required in the airgap if a proper
calculation of maxwell stresses is to be performed. If instead a volume
integral of JxB is performed, two layers are enough. With only a few
exceptions, we have used four layers for both methods.

Area close to air

gap with quadratic
elements )LJ

71
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

7.4.5 Influence on mesh size.

)LJ'(GG\
FXUUHQWGLVWULEXWLRQ
IURPDFRDUVHPHVK

Compare fig. 45 above with fig. 49. The model in fig. 49 has 51.839
equations. All calculations converge and the plotted current distribution is
smooth. In fig. 45 the mesh is much too coarse. Throughout
this report we have been using between 100.000 and
150.000 equations for end effect calculations and
about 50.000 to 60.000 equations for the
center magnets. The copper cylinder is
rotating with no eccentricity. Thus
there can be no induced eddy
currents. However, due to the
discrete non-circular mesh eddy
currents with no physical
significance are induced.
Three-dimensional arrows
show the direction of the
eddy currents.

)LJ7KLVPHVKLVWRRFRDUVH
DQG UHVXOWV LQ EDG FXUUHQW
GLVWULEXWLRQVOLNHLQILJ

72
 0HVKJHQHUDWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

False eddy currents

Non-circular air gap

Too flat elements

)LJ

Fig. 47 shows the false eddy currents induced solely by the bad mesh. The air
gap elements are too wide resulting in a non-circular mesh. In fact the mesh
is always non-circular, and we found a quadratic relationship between the
width of the elements and the mesh related fictitious losses.

A good compromise turned out to be to choose the

element width to 1.25 degrees, as mentioned
earlier. In Mega, using the “arc of
element” command where 288
elements are specified around the
full circumference does this.

The triangular elements

connecting the arcs of
different width have to be
made by hand in order to
make them symmetric,
otherwise results as in
fig. 48 can be obtained.

)LJ$V\PPHWULFPHVK

73
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 6RXUFHVDQGUHJLRQV
In Mega each mesh element is given certain material properties like
conductivity and permeability. Materials are referred to as regions. Also
properties useful for permanent magnets like remanence and direction of
magnetization can be applied. Thus there is no need for current sources and
surface currents when modelling permanent magnets.

Thanks to the Minkowski transform speed is treated as a region property, so

no time step and no sliding interfaces are needed.

The following regions have been used:


&RSSHUF\OLQGHU

σ = 60 M(Ωm)-1
n = 9 000 – 600 000 RPM

,URQSROHVKRH

µr = 300 (or nonlinear construction steel)

3RZGHUSROHVKRH

µr = 1 – 500

)HUULWHSROHVKRH

µr = 1 000 – 40 000

0DJQHWV

µr =1.0 or 1.05 (Flywheel bearing)

Br=1.0 magnetized in cartesian z- direction. Flywheel radial magnets are
radially magnetized.

'LVWDQFHULQJV

µr =1.0
σ = 0 M(Ωm)-1

74
 (GG\FXUUHQWGLVWULEXWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 (GG\FXUUHQWGLVWULEXWLRQ
When the bearing is operating in an eccentric mode eddy currents are induced
in the conducting cylinder, as has been explained in chapter 6. However,
since the conductor is a homogenous bulk cylinder, the simplified eddy
current paths described in chapter 6 has to be modified to a more continuous
current distribution. Fig. 49 below shows the modulus of the current densities
on the surface of a single row intermediate bearing.

)LJ

The tangential currents in the middle, close to the magnets and colored red
and yellow, represents a current density of 4 – 5 A/mm2. The return currents
on the sides are blue representing approximately half that value.

Inside the cylinder the current density will be highest on the surface close to
the magnets and decrease inside the material, fig 50a. This can be even better
viewed in the flywheel bearing in fig. 50b.

75
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

Especially the flywheel bearing illustrates this since that particular bearing
has a rotor cylinder that is very thick. The currents have no chance to
penetrate the whole cylinder. There are two reasons for the current decay
inside the cylinder:

• Exponential decay of radial flux due to skin effect.

• Geometric decay of radial air gap flux density due to the pole
design.

)LJDE

Thus when designing the conducting cylinder there is no meaning in having a

cylinder wall that is much thicker than one skin depth. The skin depth should
of course be calculated for the lowest operating speed, since the skin depth is
larger at low speed. In case both inner and outer magnets are used, the wall
thickness may be as high as 2 times the skin depth, since the flux is
penetrating from both sides.

In chapter 6 we derived an expression for the current density based on the

flux derivatives. In order to give a complementary and very simple
explanation of the direction of the main eddy currents, neglecting the stray
ones, we can use Ampere´s model of a permanent magnet combined with
Lentz law. According to Ampere the magnet can be described as covered
with a current sheet around the poles where the surface current density Ji
equals the intrinsic remanens of the magnet divided by µ0, se fig. 51a. The
sheet closest to the conducting surface will have the most influence on the
induced currents, and according to Lentz law a voltage will be induced that
will drive a current so as to counteract any change of flux. Thus the direction
will be opposite, that is repulsive, with respect to this current sheet when that
part of the conducting cylinder is closing up to the magnet, and equally
directed, that is attractive, when that part of the cylinder is moving away

76
 (GG\FXUUHQWGLVWULEXWLRQ
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

from the magnet, fig.51b. Thus the currents will try to simultaneously push
and pull the rotor back to the center whenever it is displaced.

Now this is not enough to explain the current distribution. We also need to
take the phase delay due to inductance in the magnetic circuit into
consideration. This results in a speed dependent current phase shift, as shown
in fig. 28 a-b and 52 a-c. The eddy currents want to oppose the change in
flux, which occurs within the range of 0 to 180 degrees before the angle of
the displacement. The phase shift varies from close to 0 degrees at low speed
to 90 degrees at infinitely high speed. Thus at very high speed the currents
occurs within a range from 90 degrees before to 90 degrees after the angle of
displacement. In this case the force is directed parallel to and opposite to the
direction of the displacement.

In fig. 51a-c the speed is high, 90 000 RPM, and the force is almost parallel.
Using the Ampére surface currents in fig. 51a and comparing it with the real
currents in fig. 51b it is easy to understand how repulsive and attractive
forces occur in the bearing. This model also helps understand which
parameters that are important when optimizing the bearing.

Equivalent Eddy currents

surface
currents

\

]
[

)LJDF&XUUHQWVLQDQLQWHUPHGLDWHEHDULQJ

7.6.1 Field diffusion

The phase shift is of great importance for the force distribution and for
bearing stability. Thus we will perform many parametric studies to see which
variables influence the phase shift, and discuss to what extent they can be
optimized. But we can make some conclusion directly from fig. 51. Compare
fig. b and c. The speed and the eddy currents are the same, as can be seen
from the colors. But we have chosen two different subset of elements in
which we display the current vectors. In fig. 51b the middle of the cylinder is
chosen, and in fig. 51c the inner surface has been chosen. There is a clear
phase shift between the eddy currents in the middle and the ones on the
surface, to the benefit of the former, since we want the phase shift to be as
large as possible. This is due to the time it takes for the currents to penetrate

77
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

)LJ)O\ZKHHOEHDULQJ)LHOGGLIIXVLRQLOOXVWUDWHGE\TXDVL
VWDWLRQDU\DFFHOHUDWLRQ

the conducting material. Thus it is obvious that in order to decrease the force
angle we should use a thick cylinder. However, we concluded above that it is
no use to have it much thicker than one skin depth, so from the stability point
of view we have already fixed one variable.

In fig. 52 a flywheel bearing is analyzed at three different speeds. To the left

the speed is only 15 000 RPM and the current density is low. In the picture in
the middle the speed is increased to 45 000 RPM and the currents are much
larger. But there is also a phase delay, which increases with increasing radius.
At high speed, 90 000 RPM, this effect is very dominant, as can be seen to
the right. This effect is called field diffusion, and these pictures beautifully
illustrate the well known electrodynamic field diffusion equation.

7.6.2 Importance of flux derivatives

To fully understand the bearing we have to emphasize the importance of flux
derivatives. It must be remembered that the eddy currents are generated by
time derivatives of the magnetic flux as seen from the rotor, and not by the
flux itself. A strong flux does not guarantee large eddy currents, and vice
versa there can still be large eddy currents in the center of the cylinder even if
this part is operating in a null flux region. Such a region can be generated by
counteracting magnets acting on both sides of the cylinder as in the
intermediate bearing arrangement presented in fig. 51. The intermediate
bearing has the lowest radial flux but as we will see it has the largest stiffness
of the three bearing types presented in this report. This is because it has the
largest radial flux derivatives.

Finally, a last important aspect will be given before we start calculating the
forces and doing some parametric studies: Large time derivatives are caused
by large radial space derivatives, which requires large changes over small
distances. But above we concluded that for good stability we should use thick
walled cylinders to reduce resistance, which is contradictory to the first
demand. Thus we can expect that we can never optimize a bearing for both
maximum stiffness and maximum stability at the same time.

78
 %HDULQJIRUFHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 %HDULQJIRUFHV
In Mega there are basically four methods to calculate forces. They all have
some advantages and disadvantages. They are:

1. Volume integral of Lorenz forces

2. Surface integral of Maxwell stresses
3. Magnetic potential energy difference
4. Volume integral of Joule heat

The volume integral of Lorenz forces is very accurate since it involves a lot
of elements. It does though become less accurate at very high speed due to
the fact that the skin effect reduces the number of elements carrying
relatively large current, as can be seen in fig. 53 where these elements are
yellow. The surface integral of Maxwell stresses does not have this problem,
since the air gap surface area is constant at all speeds. The disadvantage of
Maxwell stress is that for a bearing with more than one air gap, like the
intermediate bearing, the surface integral has to be calculated twice in the
post processor. The volume integral does only have to be calculated once.

If ferromagnetic materials are used in the rotor, Maxwell stresses have to be

used, since Lawrence forces does not take the attractive and destabilizing
force between iron and magnets into consideration. Many such calculations
were performed, but in this dissertation we will focus on bulk rotors without
iron parts.

Energy methods are sometimes useful, but only a few were performed since
they require two calculations for each value, and the result obtained is due to
severe cancellation.

Joule heat can be used to calculate brake torque and bearing losses, and thus
equals the mechanical losses in the bearing. Thus

1
σ ∫ - 2 G9 = 0 ⋅ ω = 3
] ORVV

Of the four methods Joule heat is the one most sensitive to mesh accuracy.
The reason is that a non-circular mesh will induce additional stray eddy
currents. These currents will introduce forces and losses, but the Joule heat
losses are squared and thus always positive, while the torque calculated by
method 1 and 2 contains both positive and negative contributions, which are
summed and thus cancelled. Thus losses calculated using method 4 are
always too high. We did not use this method very much, just for comparison.
It is though possible to calibrate the result by subtracting the losses calculated
when the rotor is running in the center position, see fig. 55. But this also
requires two calculations, which we want to avoid. The Joule heat method

79
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

can not be used to calculate the restoring force in the y-direction, )y, but it
can actually be used to calculate the side force in the x-direction )x, provided
that the flux is truly homopolar. In the latter case 0z and )x are related as:

0 ] = \)[ = − Jε)[

To conclude, all methods were used for comparison until the mesh was
sufficiently good and the values converged. During the rest of the study only
the Lorenz forces were used, since they provide the fastest method, and yet
gave as accurate results as the Maxwell stress method once the mesh was
acceptable. Fig. 54 and 55 show a comparison between the results from 3 of
the methods.

To view the forces in Mega, it is possible to select a subset of elements and

draw the forces as three-dimensional arrows as in fig. 53. The colors
represent the current densities as before. It can be seen that in the areas with
largest current densities, yellow areas, the forces are the largest, as expected.

)LJD530 )LJE530

)LJF530 )LJG530

80
 %HDULQJIRUFHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

/LIWDQG6LGH)RUFHV

4 &DOFXODWLRQPHWKRG

@3 Fy, Maxwell stress

>1
H Fy, Lorenz force
FU Fx, Maxwell stress
R
)2 Fx, Lorenz force

0
0 0,05 0,1 0,15 0,2
'LVSODFHPHQW>PP@

)LJ/LIWIRUFH)\DQGVLGHIRUFH)[FDOFXODWHGXVLQJ
GLIIHUHQW PHWKRGV 6DPH EHDULQJ DV LQ ILJ  ZLWKRXW
SROHVKRHV530

%UDNH7RUTXH

3,0E-04

@ &DOFXODWLRQPHWKRG
P
1
>H
2,0E-04
Mz, Maxwell stress
XT
UR Mz, Joule heat*
W
HN 1,0E-04 Mz, Lorenz force

UD% * Calibrated at (0,0)

0,0E+00
0 0,05 0,1 0,15 0,2
'LVSODFHPHQW>PP@

)LJ7RUTXHFDOFXODWHGDW530XVLQJWKUHHPHWKRGV

81
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

The forces we want to calculate with Mega are the restoring lift force in the
y-direction and the side force in the x-direction. We also simulate the braking
torque to cross check it with the force x-component. From these values we
then can derive the in plane stiffness, the force angle and finally the losses.

Let us analyze fig. 53a little bit closer. Take a look at the direction of the
forces. Comparing fig. 53c and d with 51b and c we see that the forces in
each element are always perpendicular to the currents. Further, the direction
changes gradually as speed increases. The x-component is relatively large at
low speed and decreases drastically at high speed. (It should be noted that in
fig. 53 the colors and arrows are recalibrated for each new picture, so they
can´t be used to compare the actual values of the forces or current densities
between two pictures. However, in fig. 52 the same scaling parameters are
used to enable comparison)

At 90 000 RPM it seems like the force is perfectly restoring and that the force
angle is zero, but this is true only for the elements in the middle of the
conductor, fig. 53c. In fig. 53d the speed is the same, but the subset of
elements is chosen close to the rotor surface where the phase delay is less.
Here it is clear that we still have a considerable x-component of the force.
The volume integral
) = ∫ - × % G9

results in the average of the Lorenz forces over all elements in the cylinder.
The average will always contain a certain x-component, which we will try to
limit by finding a proper bearing design.

From fig. 53 it is clear that speed is a very important variable in determining

the side force. Also restoring forces are speed dependent, as we know from
chapter 6. At low speed there are almost no forces at all. Let us continue our
study by plotting these forces versus speed for the intermediate-bearing
model in fig. 53. A principle sketch of this type of bearing can be found in
fig. 22. Rotor displacement is 50 µm in the negative \-direction. The results
can be found in fig. 56 to 60 and the bearing data are as follows:

'HVFULSWLRQ Intermediate bearing with (without) pole shoes

0DJQHWV 2 outer ring magnets. 38 × 30 × 4 mm

2 inner ring magnets 24 × 16 × 4 mm

3ROHVKRHV 1 outer intermediate washer 38 × 30 × 2 mm

(when used) 2 outer end plates 38 × 30 × 1 mm
1 inner intermediate washer 24 × 16 × 2 mm
2 inner end plates 24 × 16 × 1 mm

5RWRU Copper cylinder 29 × 25 × 9 mm

82
 %HDULQJIRUFHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

%HDULQJIRUFHV
DWFRQVWDQWHFFHQWULFLW\H 

2,5

2,0

@ 1,5
1
>H 5RZ%HDULQJ Fy
FU
R) 1,0 ZLWKRXWSROHVKRHV Fx

0,5

0,0
0 200000 400000 600000
6SHHG>530@

)LJ

%HDULQJIRUFHV
DWFRQVWDQWHFFHQWULFLW\H 

1,2

1,0

0,8
@
1
>H 5RZ%HDULQJ Fy
FU 0,6 ZLWKSROHVKRHV Fx
R
)
0,4

0,2

0,0
0 200000 400000 600000
6SHHG>530@

)LJ

83
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

)RUFHUHVXOWDQW
DWFRQVWDQWHFFHQWULFLW\H 
Without pole shoes With pole shoes
2,5

2,0
@
>1
W
Q 1,5
WDO
X
VH
U 1,0
HF
UR
)
0,5 5RZ%HDULQJV

0,0
0 200000 400000 600000
6SHHG>530@

)LJ

)RUFHDQJOH
Without pole shoes With pole shoes

90

80

70
@ 60
J
H
G
> 50
HO
J
Q
D 40 5RZ%HDULQJV
HF
U 30
R
)
20

10

0
0 200000 400000 600000
6SHHG>530@

)LJ

84
 %HDULQJIRUFHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

6WLIIQHVV
Without pole shoes With pole shoes

60

50

@ 40
P
P

1
>V 30
VH
QII
WL 20
6
10 5RZ%HDULQJV

0
0 200000 400000 600000
6SHHG>530@

)LJ

In fig. 54 to 57 we have studied how the x and y- components of the bearing

force and the losses relates to displacement and to speed. In the speed
calculations the displacement was set to 0.05 mm, which corresponds to an
eccentricity H of 10%. We are now able to make some conclusions:

1. )[ and )\ are linear functions of displacement, as predicted.

2. 0] is a quadratic function of displacement.
3. )[ increases linearly at low speed and decreases asymptotically
towards zero at high speed.
4. )\ increases asymptotically to a certain finite value at high speed.
5. Iron pole shoes reduce )[ and )\.
6. Iron pole shoes have a positive effect on force angle.

In fig. 58, we have derived the resulting bearing force, using the results from
fig. 56 and 57. Thus, these values are also valid for an eccentricity of 10%.
The corresponding force angle is found in fig. 59.

It is unpractical to use eccentricity dependent quantities. Since the forces are

linear with displacement, it means that their relationship expressed as force
angle, is constant. Also stiffness is constant since space derivatives of linear
functions are constant. Thus neither force angle nor stiffness is dependent on
eccentricity. Henceforth we will use these quantities instead of forces. The
stiffness curve in fig. 60 is identical to the fig. 58, except for the units.

85
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 %HDULQJORVVHV
Losses can be calculated directly in the Mega post processor using the
volume integral of Joule heat, but due to disadvantages with this method as
described in chapter 7.1 losses will instead be calculated from the brake
torque. The torque will be calculated as the integral of the Lorenz forces.

In fig. 61 the torque is plotted as a function of speed. The same bearing as in

fig. 54 to 60 is used, and it is clear that using pole shoes has a large effect on
the torque. It not only reduces the torque, it also moves the maximum torque
towards lower speeds, which is to be expected with increasing inductance.

Multiplying torque with angular speed is a simple operation, which results in

bearing losses for that particular eccentricity, fig. 62. However, we do not
have much practical use of this graph. Instead it is much more interesting to
study the bearing at constant load, fig. 63. This curve is derived by dividing
the load with the actual stiffnes from fig. 60 resulting in the actual
eccentricity for that particular speed. Thereafter losses are quadratically
extrapolated from the losses in fig. 62. The dotted line in fig. 63 represents
the emergency bearing, which, according to some often used substandard in
magnetic bearing technology, should prevent eccentricity from exceeding
50%.

%UDNHWRUTXHDWFRQVWDQWHFFHQWULFLW\H 
Without pole shoes With pole shoes

6,0E-05

5,0E-05

@
P
>1 4,0E-05
HX
TU 3,0E-05
WR
JQ
LN 5RZ%HDULQJV
DH 2,0E-05
U%
1,0E-05

0,0E+00
0 200000 400000 600000
6SHHG>530@

)LJ

86
 %HDULQJORVVHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

3RZHUORVVDWFRQVWDQWHFFHQWULFLW\H 
Without pole shoes With pole shoes

2,5

2,0

@
>:
V 1,5
VR
O 5RZ%HDULQJV
UH
Z
R
1,0
3
0,5

0,0
0 200000 400000 600000
6SHHG>530@

)LJ

3RZHUORVVDWFRQVWDQWORDG

5RZ%HDULQJZLWK
SROHVKRHV
12

10

@ 8 5N
>:
V H  4N
VR
O 6 3N
HU 2N
Z
R 4 H!
3 H 1N

0
0 200000 400000 600000
6SHHG>530@

)LJ

87
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

 3DUDPHWULFRSWLPL]DWLRQRIEHDULQJJHRPHWU\
The properties of the induction bearing are by nature strongly speed
dependent. It is thus important to evaluate all geometric changes at different
speed, since the bearing has to be able to operate over a wide speed range.
This requires a large number of calculations. In Mega rotational speed can be
calculated either by time step or by Minkowskij transformation. We have
chosen the second alternative since it is the most time saving method.

If high stiffness alone would be the only bearing requirement it could easily
be achieved by using large magnets and by always operating the bearing at
high speed. Reality is not that simple and we have formulated the following
goals for the parametric optimization.

Our goal is to find a bearing geometry that

1 works well over a wide speed range with regard to stiffness,

losses and stability
2 reduces the amount of magnet material required for a certain
stiffness
3 has a design well suited for mass production.

We will start the optimization process by studying how scaling effects

bearing properties. After that additional magnets will be added to improve
stiffness. The magnets and poles will be optimized with regard to geometric
parameters and material properties. Finally leakage flux will be analyzed and
prevented.

88
 3DUDPHWULFRSWLPL]DWLRQRIEHDULQJJHRPHWU\
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

7.9.1 Scaling
Scaling of magnetic properties is straightforward, and very useful when the
operating range has to be moved or extended in any direction. Increasing the
bearing size with the linear scaling factor s, affects the bearing properties
proportionally to the air gap surface area as follows:

1
Magnetic forces: )1 = )0 V 2 if ω1 = ω 0
V2
1
Magnetic torque: 71 = 70 V 2 if ω1 = ω 0 2
V
1
Stiffness: .1 = . 0 V if ω1 = ω 0 2
V
σ0 1
Speed: ω1 = ω 0 ⋅ for a change in con-
σ1 V2 ductivity from σ0 to σ1.

Thus doubling the size will increase the forces a factor of four, if the skin
depth has increased proportionally. This in turn requires the speed, (or the
conductivity) to be reduced a factor of four.

The magnet material and thus cost, will increase with the volume, which
make it a factor of eight.

Temperature affects the remanence of the magnets and the conductivity. For
a given bearing that is used in a cryogenic system at a temperature of 77 K
instead of at room temperature, the conductivity is roughly 16 times higher
and we can expect the same forces at one fourth of the speed due to the
conductivity, and some additional force due to the increased properties of the
magnets. Thus, this bearing is ideal in cryogenic environments.

Centrifugal forces on a rotor, for instance a flywheel, scales to the square of

the peripheral speed. Thus, when scaling a rotor that already is limited by
mechanical stress, speed has to be inversely proportional to the scaling factor.
Doubling the size requires speed to be one half. But above we have seen that
we can reduce speed a factor of four. Reducing speed only a factor of two
will move us further to the right in the speed diagram, which is advantageous
for the stiffness and the force angle. Thus scaling up a machine has a positive
effect on bearing performance.

89
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

7.9.2 Adding intermediate magnets

Adding an extra magnet to the bearing will of course increase stiffness.
However, we have already reached the calculation limit for our FEM
program MEGA, so in order to estimate the effect of this additional magnet
we have to make a new model of that extra magnet alone, see cut A-A to B-B
in fig. 41. This result is then added to our data for the 2-row bearing.

By using proper symmetry conditions as described in chapter 7.1 we can

expect good agreement with reality for bearings with many magnets.
However, there will be an error due to the fact that the current normal to
plane A-A may not be the same on both sides of the plane, since we are using
different models for the two sides. It is likely that the current is less in the 2-
row bearing since it has more flux leakage due to end effects. This error will
be most accentuated in 3-row bearings, and will not have any significant
effect for bearings with large number of magnets.

Now look at fig. 64. The stiffness from the 2-row bearing is recognized from
fig. 60. The second curve in fig. 64 is the result from one additional magnet
row. This “row” consists of one outer and one inner magnet. (The term
“magnet pair” would be more appropriate, but we have chosen to use the
same vocabulary as for ball bearings.) Observe that this single row has about
the same stiffness as the whole bearing consisting of 2 rows of magnets!

6WLIIQHVV

25

20

@
P15 2-Row Bearing
P

>1
V
VH
QII 10 Model of one additional
WL Intermediate Magnet Row
6
5

0
0 200000 400000 600000
6SHHG>530@

)LJ

90
 3DUDPHWULFRSWLPL]DWLRQRIEHDULQJJHRPHWU\
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

7.9.3 Optimization of magnet length

In the previous subsection we noted that the intermediate magnets were
double as effective as the end magnets. Thus it is obvious that induction
bearings shall have many rows and that the optimization work shall mainly
focus on these magnets.

A realistic design has to take geometric restraints into consideration. We

can´t have too many magnets in a high-speed machine since rotor length is
critical due to rotor dynamics. Thus we have to choose between either a few
axially long magnets, or many shorter magnets to limit the total bearing
length.

In fig. 65 stiffness has been plotted versus the axial length of an intermediate
magnet pair. The calculations have been performed for two different speeds.
There is an optimum around 6 mm at high speed and 9 mm at low speed,
when axial restraints are not taken into consideration.

In fig. 66 the results have been divided by the additional bearing length,
including the magnet and one additional pole shoe. Here the optimum has
changed and the length should be 3 mm at 90 000 RPM and 5 mm at 30 000
RPM, for highest stiffness.

6WLIIQHVVSHUPDJQHWSDLU

25

20

@
P15
P
1
>V 90,000 RPM
VH 30,000 RPM
QII 10
WL
6

0
0 2 4 6 8 10 12
0DJQHWOHQJWK>PP@

)LJ

91
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

6WLIIQHVVSHUD[LDOOHQJWKXQLW

2,5

 @
2
P
P

1
>
WKJ 1,5
QH
OO 90,000 RPM
DL 30,000 RPM
[D
U 1
HS
V
VH
QII
LW 0,5
6

0
0 2 4 6 8 10 12
0DJQHWOHQJWK>PP@

)LJ

)RUFHDQJOH

90

80

70

@J 60
HG
> 50
OHJ 30,000 RPM
QD 40 90,000 RPM
H
FU
R) 30

20

10

0
0 2 4 6 8 10 12
0DJQHWOHQJWK>PP@

)LJ

92
 3DUDPHWULFRSWLPL]DWLRQRIEHDULQJJHRPHWU\
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

In fig. 67 the force angle is shown. If rotordynamic stability is a concern, for

instance when damping is difficult to achieve, the force angle is likely the
most important parameter to optimize. Long magnets should definitely be
used in this case.

For both stiffness and force angle speed is the most important parameter.
Thus, it is important to know the operating speed before we can fully
optimize the bearing.

In our examples this far we have chosen magnets that are 4 mm long, which
is a good compromise between low- and high-speed stiffness. It also has a
fairly good force angle at high speed.

With this magnet in mind, we are ready to sum our results and study the
properties of magnets with many rows, fig. 68-71. All bearings are equipped
with pole shoes to decrease the force angle.

In these diagrams, bearings with up to 4 pairs of magnets are plotted. It is

trivial to estimate the stiffness for bearings with more magnets simply by
summing, and the force angle does not change very much when more
magnets are added.

Constant load curves like fig. 63 can be derived for any number of magnets
by using the results in fig. 68 and 71, as explained earlier.

A little note should be given to the 1-row bearing. Especially from fig. 61 it
is clear that it does not quite seem to follow the pattern. The one row bearing
is calculated using a separate model. It differs from the other models in that
the rotor is considerably longer than the stator. If not, there would be no room
in the conducting rotor for eddy current return paths.

In the next subsection we shall study the somewhat complicated effects of the
conductor return paths of two row bearings.

93
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

6WLIIQHVV

80

70

60
@
P50
P
4-Row Bearing
>1
V 40 3-Row Bearing
VH 2-Row Bearing
QII
LW 30 1-Row Bearing
6
20

10

0
0 200000 400000 600000
6SHHG>530@

)LJ

)RUFHDQJOH

90

80

70

@J 60
HG 1-Row Bearing
>H 50
OJ 4-Row Bearing
QD 40 3-Row Bearing
H
FU 2-Row Bearing
R) 30
20

10

0
0 200000 400000 600000
6SHHG>530@

)LJ

94
 3DUDPHWULFRSWLPL]DWLRQRIEHDULQJJHRPHWU\
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

%UDNHWRUTXHDWFRQVWDQWHFFHQWULFLW\H 
$OOEHDULQJVZLWKSROHVKRHV
9,0E-05

8,0E-05

7,0E-05
@
P
>1
6,0E-05
HX 5,0E-05 4-Row Bearing
TU 3-Row Bearing
WR 4,0E-05 2-Row Bearing
JQ
LN 1-Row Bearing
DH 3,0E-05
U%
2,0E-05

1,0E-05

0,0E+00
0 200000 400000 600000
6SHHG>530@

)LJ

3RZHUORVVDWFRQVWDQWHFFHQWULFLW\H 
$OOEHDULQJVZLWKSROHVKRHV
2,0

1,5

@
>:
V 4-Row Bearing
VR 3-Row Bearing
O 1,0
UH 2-Row Bearing
Z
R 1-Row Bearing
3
0,5

0,0
0 200000 400000 600000
6SHHG>530@

)LJ

95
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

7.9.4 Optimization of rotor length

We have now optimized the length of the stator, and an interesting question is
whether the conducting rotor should be equally long, or if there is a better
choice. Fig. 74 shows stiffness plotted against relative rotor length for a 2-
row bearing spinning at 90 000 RPM. For corresponding losses, see fig. 75.

At first the curves may seem rather confusing, but it is made quite clear when
we take a look at the eddy current paths in fig. 72.

)LJ/RQJURWRUWRWKHOHIWDQGVKRUWURWRUWRWKHULJKW6WDWRUSDUWVGDVKHG

In the short rotor to the right there is only place enough for the main eddy
current. It has no choice, but is forced to circulate close to the magnets where
it can produce a large force. But the resistance is
comparatively high, since the conductor volume is
limited. The long rotor to the left has space enough
for stray eddy currents, which help in increasing
the active current under the magnets. The worst
case, fig. 73, at least when pole shoes are used, is
when the rotor is equally long as the stator. There
are small and narrow stray currents, and they are
situated close to the end plates where they produce
negative stiffness, and thus reduce overall bearing
performance. Thus, the conducting rotor cylinder
should be either somewhat shorter, or much longer
than the stator.

)LJ

96
 3DUDPHWULFRSWLPL]DWLRQRIEHDULQJJHRPHWU\
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

6WLIIQHVVYHUVXVUHODWLYHURWRUOHQJWK
With pole shoes Without pole shoes

35,0

30,0

@ 25,0
P
P20,0
1
>V
VH 15,0
QII
LW
610,0 URZEHDULQJV

5,0
530
0,0
0,25 0,5 0,75 1 1,25 1,5 1,75 2
/ /
UR VW

)LJ

)RUFHDQJOHYHUVXVUHODWLYHURWRUOHQJWK
With pole shoes Without pole shoes

80,0

70,0

60,0
@J
HG 50,0
>
OHJ 40,0
Q
D
HF 30,0
UR
20,0 ) URZEHDULQJV

10,0
530
0,0
0,25 0,5 0,75 1 1,25 1,5 1,75 2
/ /
UR VW

)LJ

97
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

7.9.5 Optimization of magnet width

We shall now study how the cross-section width of the magnets affects the
stiffness and force angle. Consider the following bearing:

'HVFULSWLRQ Intermediate bearing without pole shoes

0DJQHWV 2 outer ring magnets. 'om × 59 × 8 mm

2 inner ring magnets 49 × Gim × 8 mm

5RWRU Copper cylinder 58 × 50 × 18 mm

In fig. 76-78 bearing data are found for two different speeds. The diagrams
are plotted against the magnet cross-section width to length ratio, ZO. Thus
ZO 1 means that the magnets has a square cross section of 8 × 8 mm, and
that 'om = 75 mm andGim = 33 mm. As expected stiffness increases fast with
increased width, but from fig. 78 it can be seen that it is not very economical
to increase width too much. For this bearing, the optimum ratio, from an
economical point of view, is ZO 0,5.

If higher stiffness is required, the designer has the choice between either
more magnets, which is the most economic choice, or wider magnets, which
is better if rotor length is critical.

6WLIIQHVV

160,00

140,00

120,00
@
P100,00
P
1
>V 80,00 90 000 RPM
VH 30 000 RPM
QII 60,00
WL
6
40,00

20,00

0,00
0 0,2 0,4 0,6 0,8 1 1,2 1,4
ZO

)LJ

98
 3DUDPHWULFRSWLPL]DWLRQRIEHDULQJJHRPHWU\
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

6WLIIQHVVSHUFPPDJQHWYROXPH

 @
P
F
P4
P
1>
VV 90 000 RPM
HQ 30 000 RPM
ILI
WV
H 2
YWL
ODH
5

0
0 0,2 0,4 0,6 0,8 1 1,2 1,4
ZO

)LJ

)RUFH$QJOH

50,00

40,00

@J
HG 30,00
>
OHJ 30 000 RPM
Q
D 90 000 RPM
HF 20,00
UR
)
10,00

0,00
0 0,2 0,4 0,6 0,8 1 1,2 1,4
ZO

)LJ

99
)(0VLPXODWLRQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

7.9.6 Leakage flux reduction

The radial flux in the iron pole shoes does unfortunately leak to the outside of
the bearing as well as through the conducting rotor, fig. 23. A way to reduce
this leakage is to reduce the size of the pole shoe washer on the side where no
leakage is desired, as has been done to some extent in fig. 23. This will
increase the leakage flux in the air gap.

Which diameter of the iron washer is optimal? Let us study a flywheel

bearing with the following data:

'HVFULSWLRQ Flywheel bearing with pole shoes

0DJQHWV 4 inner axial ring magnets 24 × 10 × 7 mm

3ROHVKRHV 3 inner intermediate washer 24 × diw × 3.50 mm

2 inner end plates 24 × diw × 1.75 mm

5RWRU Copper cylinder 35 × 25 × 37,5 mm

We shall now optimize the inner diameter diw of the iron washer and replace
it with a non-magnetic distance ring as shown in fig. 29. The results are
shown as dashed lines in fig. 79-80, and we can conclude that 13 mm would
be the best choice. If we chose 24 mm, which is the same as not having any
iron pole shoe at all, the stiffness is high, but the force angle is dramatically
increased due to the reduced inductance.

6WLIIQHVV
5RZ)O\ZKHHO%HDULQJ
100
@
P 80
P
 With radial
1
>V 60 magnets
VH 40 Without radial
QII magnets
LW 20
6
0 530
10 12 14 16 18 20 22 24
,URQZDVKHULQQHUGLDPHWHU>PP@

)LJ

100
 3DUDPHWULFRSWLPL]DWLRQRIEHDULQJJHRPHWU\
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

For comparison we shall also replace the distance rings for radially
magnetized ring magnets, the direction of magnetization being opposite to
the leakage flux. This is very efficient to reduce the leakage flux and to
increase stiffness. Unfortunately such magnets are very expensive and are not
yet suited for mass production.

If they will ever reach the same price level as axially oriented magnets, fig.
81 is useful for optimization with regard to cost. A good compromise with
regard to stiffness, cost and stability would be to choose the diameter in the
interval between 16 and 18 mm.

)RUFH$QJOH
5RZ)O\ZKHHO%HDULQJ
35
@
JH
G> 30 With radial
OH magnets
J 25
QD Without radial
H
FU 20 magnets
R
) 15 530
10 12 14 16 18 20 22 24
,URQZDVKHULQQHUGLDPHWHU>PP@

)LJ

6WLIIQHVVSHUFPPDJQHW
5RZ)O\ZKHHO%HDULQJ
V 6
VH @
QI P
IWL F

VH P5 With radial magnets
LYW P
DO 1

H > 530
5
4
10 12 14 16 18 20 22 24
,URQZDVKHULQQHUGLDPHWHU>PP@

)LJ

101
([SHULPHQWDOYDOLGDWLRQRIWKHSURWRW\SHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB



8 Experimental validation of
the prototypes
 7KHEHDULQJSURWRW\SHV
To accurately verify the FEM-simulations, a test rig with a fully levitating
rotor is being manufactured, and the results will be presented in a follow up
report. A special motor7 has been developed that does not produce any
disturbing radial forces.

However, before designing the test rig we wanted to get some field
experience from the bearing to get a feeling for rotordynamic stability and
damping requirements. It was decided to manufacture such a test rig by
modifying an existing turbomolecular vacuum pump from Balzers Pfeiffer
GmBH. Below we will give a short presentation of this very simple test and
discuss some of the results we obtained.

The pump chosen was a TMH 071 turbomolecular vacuum pump. These
pumps are originally fitted with one ball bearing and one passive radial
magnetic bearing. It also has one auxiliary ball bearing on the side of the
magnetic bearing.

Modifying the passive bearing into an induction bearing was quickly done,
and in order to improve low speed stability one of the rotor magnets was not
removed. The other two were exchanged for a copper cylinder. See fig. 82
and 83. All original stator magnets were kept unchanged.

Vibration damping on the original vacuum pump is done on the ball bearing
side. There is no damping on the magnetic bearing side. According to our
analysis damping is necessary, and we were not certain that the damping on
the ball bearing side would be enough. Unfortunately there was definitely no
space over for an additional damper on the side of the magnetic bearing.

To reduce the need for additional damping it would be advantageous to use

long magnets, as can be seen in the diagram in fig. 67. However, we had to
keep the original magnets on the stator due to space limitations. Thus we had
a bad design and were not sure it would work, so it was decided that the
pump should have an additional explosion protection, and it should also be
remotely controlled in case of any failure at 90.000 RPM.

The ball bearing in the lower end of the shaft was unchanged. We also used
the original motor, which is a permanent magnet brushless DC motor.

102
 7KHEHDULQJSURWRW\SHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

The following changes were made:

• The turbine blades were taken away to reduce air drag losses.
• The stator blades were removed and replaced by a protective
cylinder inside the vacuum housing.
• Three eddy current sensors were mounted through the vacuum
chamber with vacuum seals.
• Two of the rotor magnets were replaced by a copper cylinder.
• For some experiments thin iron or aluminum washers were inserted
between the stator magnets.

In addition a tilt- and vibration table was manufactured, including a remote

controller.

Stator magnets

)LJ7KHLQGXFWLRQEHDULQJVWDWRU

103
([SHULPHQWDOYDOLGDWLRQRIWKHSURWRW\SHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

Rotor magnet

Copper cylinder

)LJ7KHLQGXFWLRQEHDULQJFRSSHUURWRUPRXQWHGLQVLGHWKHKROORZVKDIW

 7HVWUXQV
The pump was mounted on a vibration table and accelerated to 90.000 RPM
several times. At first in vertical position, later on tilted and finally during
fierce vibration.

The test rig was operated at a pressure of 2 mBar during the force
measurements and at 0,3 mBar during the spin down test.

The motor has an integrated variable speed drive and is able to run between
18.000 and 90.000 RPM, though the speed control was not very accurate
since it was not tuned for the low moment of inertia that was the result of
removing the blades.

Radial displacement sensors of eddy current type were attached through the
chamber with vacuum seals, fig. 86.

Radial forces were obtained by leaning the vacuum chamber in the y-

direction, fig 88a, b. This results in a speed dependent angle deviation of the
rotor in the x- and y- directions with respect to the stator. The deviations

104
 7HVWUXQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

were then measured with the displacement sensors. A third sensor, fig. 86a,
was used to measure the rotor angle for balancing purpose. The rotor was
tilted at different speeds to obtain force/displacement curves, and finally the
rotor was forced into emergency bearing contact by violent vibrations.

At first everything seemed OK, but after the pump had been dismounted
several times, it began to be more sensitive to damping.

3UREOHPV
Measuring restoring forces and side forces was supposed to be done using the
displacements sensors in fig. 86 when the pump was tilted as in fig. 88. This
however, also affected the ball bearing behaviour, and also the damping of
the pump. Thus our calibrations of the sensors were continuously altered. It is
well known that ball bearings should have a certain axial load in order for the
balls to behave properly, but since the load changed during tilting the bearing
behaviour became pretty random. The different operating modes of the balls
could clearly be heard. For this reason we did not achieve any valuable
results from the force measurements, and from the spin down tests we could
only confirm that the losses are lower than the loss-variations of the ball
bearing during its different modes.

Prototype pump

Remote pump
controllers Vibration table

Vacuum pump

Remote tilt controller


)LJ7KHWHVWULJPRXQWHGRQWKHYLEUDWLRQWDEOH6HHDOVRILJ

105
([SHULPHQWDOYDOLGDWLRQRIWKHSURWRW\SHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB


)LJDE%HDULQJVWDWRUDQGSXPSURWRUEHIRUHDQGDIWHUPRGLILFDWLRQ

106
 7HVWUXQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB


)LJDE(GG\FXUUHQWVHQVRUVDQGVHDOV

107
([SHULPHQWDOYDOLGDWLRQRIWKHSURWRW\SHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB


)LJDE&DOLEUDWLRQRIVHQVRUVZLWKPLFURPHWHUVFUHZ

108
 7HVWUXQV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

Displacement sensor

)LJDE([SHULPHQWVZLWKGLIIHUHQWUDGLDOORDG

109
&RQFOXVLRQVDQGIXWXUHZRUN
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

9 Conclusions and future work

We can conclude that the homopolar induction bearing can be analysed using
the generator model. Furthermore both the analytical model and the FEM
simulations shows that the losses are zero at centered operation. The fact that
we did not manage to measure any losses at all, even at large radial load,
indicates that this bearing has great potential for applications like flywheels
and vacuum applications were losses are of great concern.

The test rig gave some valuable information, but clearly it was not well suited
for doing measurements on, mainly due to the low damping and also due to
the unpredictable behaviour of the ball bearing.

Recommendations for future work would be to fulfill the manufacturing of

the fully levitating test rig. It would also be valuable to calculate the solution
to the diffusion equation, since this will give a more accurate analytical
solution than the generator model.

110
 5HIHUHQFHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

References
1 Earnshaw, S., “On the nature of the molecular forces which regulate
the constitution of the luminferous ether., Trans. Camb. Phil. Soc., 7,
pp 97-112 (1842)

2 Richards, P.L., Tinkham, M., “Magnetic Suspension and Propulsion

Systems for High-Speed Transportation”, J. Appl. Phys. Vol. 43, No.
6, Jun. 1972, pp. 2680-2690.

3 Sacerdoti, G., Catitti, A., Soglia, L. “ Self Centering Rotary Magnetic

Suspension Device”, US Patent No: 3811740, Publication date 1974-
05-21.

4 Lembke, T., “ Design of Magnetic Induction Bearings”, M.Sc. thesis,

Chalmers Univesity of Technology, Göteborg 1990.

5 Lembke, T., “Eddy Induced Magnetic Bearing”, European patent No.

0594033, Publication date 1994-04-27.

6 Post, R., “Dynamically stable Magnetic Suspension/Bearing System”.

US patent No. 5495221, Publication date 1996-02-27.

7 Lefevre, L., “Helical Air-gap Winding Permanent Magnet Machine”,

Electrical Machines and Drives, Royal University of Technology,
Stockholm 1998.

8 Lang, M. and Fremerey, J.K., “Optimization of Permanent-Magnet

Bearings”, 6th International Symposium on Magnetic Suspension
Technology, Turin, October 2001.

9 Filatov, A., “”Null-E” Magnetic Bearings”, Ph.D. dissertation,

University of Virginia, August 2002.

111
$SSHQGL[
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

Appendix 1
6LPSOH KDQGV RQ UXOHV IRU GHVLJQLQJ LQGXFWLRQ
EHDULQJV
Based on the analysis above we can summary the design phase of a
homopolar induction bearing into a few simple hands on rules.

&KRRVHFRQILJXUDWLRQ
Inner rotor, with ring magnets on the outside.
Outer rotor, with ring or disc magnets on the inner stator.
Intermediate rotor type, comprising a tubular rotor with magnets both on the
inside and on the outside of the rotor.
&KRRVHURWRUPDWHULDOZLWKDVKLJKFRQGXFWLYLW\DVSRVVLEOH
'HFLGHWKHORZHVWURWDWLRQDOVSHHGGXULQJFRQWLQXRXVRSHUDWLRQ
Let the wall thickness of the cylinder equal one skin depth or more for this
frequency. For intermediate rotors do not use too thick walls. A thickness of
between one and one and a half skin depth is recommended for best use of
the magnets. Choose the pole width of the magnets, including pole shoes, to
two times the sum of the airgap and the skin depth at this frequency. 
'HFLGHWKHPD[LPXPURWDWLRQDOVSHHG
Choose the maximum rotor diameter with respect to restraints like centrifugal
acceleration or other design criteria. Use bandage only on outer rotor type
bearing in order to keep the airgap small.
&KRRVHPDJQHWDUUDQJHPHQW
Halbach array will give the highest stiffness, but is expensive and need high
speed to utilize the stiffness.
Flux concentration with pole shoes gives high stiffness at low speed. Other
advantages are simple magnet arrangement, flat stiffness versus speed curve
and small side forces.
Flux concentration with conical pole shoes and magnets. Better than above
but more expensive. 
&KRRVHPDJQHWFURVVVHFWLRQGHSHQGLQJRQRSWLPL]DWLRQFULWHULD
Maximum stiffness: Let the width of the magnet equal at least two times the
pole width plus two times the axial length of a pole shoe, if the latter are
used. If disc magnets are used, the width can be exchanged for the diameter.
Thus they are more economic.
Economic stiffness: Let the width be about 60% of the value above. This will
result in maximum stiffness per kilo magnet.
&KRRVHQXPEHURIPDJQHWVIRUGHVLUHGVWLIIQHVV
Adjust by changing diameter or bearing width
&KRRVHW\SHRIEHDULQJHQG
Reduce the axial length of the end pole shoe to 50%.

112
 $SSHQGL[
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

If there is place enough, let this pole shoe cover the airgap and some of the
conducting cylinder.
Do not make the cylinder slightly longer than the stator package. Make it
either somewhat shorter or much longer.

113
$SSHQGL[
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

Appendix 2
0RGHOJHRPHWULHV

0RGHO
Description: Intermediate bearing without pole shoes

Magnets: 2 outer ring magnets. 38 × 30 × 4 mm

2 inner ring magnets 24 × 16 × 4 mm

Pole shoes: None

Rotor: Copper cylinder 29 × 25 × 9 mm

0RGHO
Description: Intermediate bearing with pole shoes

Magnets: 2 outer ring magnets. 38 × 30 × 4 mm

2 inner ring magnets 24 × 16 × 4 mm

Pole shoes: 1 outer intermediate washer 38 × 30 × 2 mm

2 outer end plates 38 × 30 × 1 mm
1 inner intermediate washer 24 × 16 × 2 mm
2 inner end plates 24 × 16 × 1 mm

Rotor: Copper cylinder 29 × 25 × 9 mm

0RGHO
Description: Flywheel outer rotor bearing with pole shoes

Magnets: 4 inner ring magnets 24 × 10 × 7 mm

Pole shoes: 3 inner intermediate washers 24 × 16 × 3,50 mm

2 inner end plates 24 × 16 × 1,75 mm

Rotor: Copper cylinder 35 × 25 × 9 mm

114
 $SSHQGL[
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

Appendix 3
3DWHQWV
The magnetic bearings presented in this dissertation are protected by the
following patents and patent applications:

US6469411 Compliant foil fluid film bearing with eddy current damper
TW471568Y Magnetic bearing
AU2285002 Compliant foil fluid film bearing with eddy current damper
WO0248561 Compliant foil fluid film bearing with eddy current damper
US6118199 Magnetic bearings
US6050782 Magnetically suspended high velocity vacuum pump
EP0960277 Magnetically suspended high velocity vacuum pump
EP0956457 Electrodynamic magnetic bearing
JP10281158 Magnetic suspension device
AU5788298 Magnetically suspended high velocity vacuum pump
AU5788198 Electrodynamic magnetic bearing
SE508445 Magnetically suspended high velocity vacuum pump
SE508442 Magnetic bearings
SE9700257 Integrating rate gyroscope
SE9700256 High velocity vacuum pump
SE9700255 Device for magnetic suspension
WO9832981 Electrodynamic magnetic bearing
WO9832973 Magnetically suspended high velocity vacuum pump

Readers interested in the patents above are kindly requested to contact the
patent department of Magnetal on the address below. For technical questions
concerning this report please contact the author directly at the University.

Magnetal Royal University of Technology

Patent Department Department of Electrical Engineering
Axel Johanssons gata 4D Electrical Machines and Power Electronics
754 50 Uppsala Teknikringen 33
Sweden 100 44 Stockholm
Sweden

115